The Algorithm Design Manual

The Algorithm Design Manual

Next: Preface Up: Main Page

The Algorithm Design Manual Steven S. Skiena Department of Computer Science State University of New York Stony Brook, NY 11794-4400 [email protected] Copyright © 1997 by Springer-Verlag, New York

● ●

●

● ● ●

Contents Techniques ❍ Introduction to Algorithms ❍ Data Structures and Sorting ❍ Breaking Problems Down ❍ Graph Algorithms ❍ Combinatorial Search and Heuristic Methods ❍ Intractable Problems and Approximations ❍ How to Design Algorithms Resources ❍ A Catalog of Algorithmic Problems ❍ Algorithmic Resources References Index About this document ...

file:///E|/BOOK/BOOK/BOOK.HTM (1 of 2) [19/1/2003 1:27:29]

The Algorithm Design Manual

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/BOOK.HTM (2 of 2) [19/1/2003 1:27:30]

Preface

Next: Acknowledgments Up: The Algorithm Design Manual Previous: The Algorithm Design Manual

Preface Most of the professional programmers that I've encountered are not well prepared to tackle algorithm design problems. This is a pity, because the techniques of algorithm design form one of the core practical technologies of computer science. Designing correct, efficient, and implementable algorithms for realworld problems is a tricky business, because the successful algorithm designer needs access to two distinct bodies of knowledge: ●

●

Techniques - Good algorithm designers understand several fundamental algorithm design techniques, including data structures, dynamic programming, depth-first search, backtracking, and heuristics. Perhaps the single most important design technique is modeling, the art of abstracting a messy real-world application into a clean problem suitable for algorithmic attack. Resources - Good algorithm designers stand on the shoulders of giants. Rather than laboring from scratch to produce a new algorithm for every task, they know how to find out what is known about a particular problem. Rather than reimplementing popular algorithms from scratch, they know where to seek existing implementations to serve as a starting point. They are familiar with a large set of basic algorithmic problems, which provides sufficient source material to model most any application.

This book is intended as a manual on algorithm design, providing access to both aspects of combinatorial algorithms technology for computer professionals and students. Thus this book looks considerably different from other books on algorithms. Why? ●

●

●

We reduce the design process to a sequence of questions to ask about the problem at hand. This provides a concrete path to take the nonexpert from an initial problem statement to a reasonable solution. Since the practical person is usually looking for a program more than an algorithm, we provide pointers to solid implementations whenever they are available. We have collected these implementations on the enclosed CD-ROM and at one central FTP/WWW site for easy retrieval. Further, we provide recommendations to make it easier to identify the correct code for the job. With these implementations available, the critical issue in algorithm design becomes properly modeling your application, more so than becoming intimate with the details of the actual algorithm. This focus permeates the entire book. Since finding out what is known about a problem can be a difficult task, we provide a catalog of important algorithmic problems as a major component of this book. By browsing through this

file:///E|/BOOK/BOOK/NODE1.HTM (1 of 3) [19/1/2003 1:27:32]

Preface

●

●

catalog, the reader can quickly identify what their problem is called, what is known about it, and how they should proceed to solve it. To aid in problem identification, we include a pair of ``before'' and ``after'' pictures for each problem, illustrating the required input and output specifications. For each problem in the catalog, we provide an honest and convincing motivation, showing how it arises in practice. If we could not find such an application, then the problem doesn't appear in this book. In practice, algorithm problems do not arise at the beginning of a large project. Rather, they typically arise as subproblems when it suddenly becomes clear that the programmer does not know how to proceed or that the current program is inadequate. To provide a better perspective on how algorithm problems arise in the real world, we include a collection of ``war stories,'' tales from our experience on real problems. The moral of these stories is that algorithm design and analysis is not just theory, but an important tool to be pulled out and used as needed.

Equally important is what we do not do in this book. We do not stress the mathematical analysis of algorithms, leaving most of the analysis as informal arguments. You will not find a single theorem anywhere in this book. Further, we do not try to be encyclopedic in our descriptions of algorithms, but only in our pointers to descriptions of algorithms. When more details are needed, the reader should follow the given references or study the cited programs. The goal of this manual is to get you going in the right direction as quickly as possible. But what is a manual without software? This book comes with a substantial electronic supplement, an ISO-9660 compatible, multiplatform CD-ROM, which can be viewed using Netscape, Microsoft Explorer, or any other WWW browser. This CD-ROM contains: ●

●

●

A complete hypertext version of the full printed book. Indeed, the extensive cross-references within the book are best followed using the hypertext version. The source code and URLs for all cited implementations, mirroring the Stony Brook Algorithm Repository WWW site. Programs in C, C++, Fortran, and Pascal are included, providing an average of four different implementations for each algorithmic problem. More than ten hours of audio lectures on the design and analysis of algorithms are provided, all keyed to the on-line lecture notes. Following these lectures provides another approach to learning algorithm design techniques. These notes are linked to an additional twenty hours of audio over the WWW. Listening to all the audio is analogous to taking a one-semester college course on algorithms!

This book is divided into two parts, techniques and resources. The former is a general guide to techniques for the design and analysis of computer algorithms. The resources section is intended for browsing and reference, and comprises the catalog of algorithmic resources, implementations, and an extensive bibliography. Altogether, this book covers material sufficient for a standard Introduction to Algorithms course, albeit

file:///E|/BOOK/BOOK/NODE1.HTM (2 of 3) [19/1/2003 1:27:32]

Preface

one stressing design over analysis. We assume the reader has completed the equivalent of a second programming course, typically titled Data Structures or Computer Science II. Textbook-oriented features include: ●

●

●

●

In addition to standard pen-and-paper exercises, this book includes ``implementation challenges'' suitable for teams or individual students. These projects and the applied focus of the text can be used to provide a new laboratory focus to the traditional algorithms course. More difficult exercises are marked by (*) or (**). ``Take-home lessons'' at the beginning of each chapter emphasize the concepts to be gained from the chapter. This book stresses design over analysis. It is suitable for both traditional lecture courses and the new ``active learning'' method, where the professor does not lecture but instead guides student groups to solve real problems. The ``war stories'' provide an appropriate introduction to the active learning method. A full set of lecture slides for teaching this course is available on the CD-ROM and via the World Wide Web, both keyed to unique on-line audio lectures covering a full-semester algorithm course. Further, a complete set of my videotaped lectures using these slides is available for interested parties. See http://www.cs.sunysb.edu/ algorith for details.

Next: Acknowledgments Up: The Algorithm Design Manual Previous: The Algorithm Design Manual Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE1.HTM (3 of 3) [19/1/2003 1:27:32]

Acknowledgments

Next: Caveat Up: The Algorithm Design Manual Previous: Preface

Acknowledgments I would like to thank several people for their concrete contributions to this project. Ricky Bradley built up the substantial infrastructure required for both the WWW site and CD-ROM in a logical and extensible manner. Zhong Li did a spectacular job drawing most of the catalog figures using xfig and entering the lecture notes that served as the foundation of Part I of this book. Frank Ruscica, Kenneth McNicholas and Dario Vlah all came up big in the pinch, redoing all the audio and helping as the completion deadline approached. Filip Bujanic, David Ecker, David Gerstl, Jim Klosowski, Ted Lin, Kostis Sagonas, Kirsten Starcher, Brian Tria, and Lei Zhao all made contributions at various stages of the project. Richard Crandall, Ron Danielson, Takis Metaxas, Dave Miller, Giri Narasimhan, and Joe Zachary all reviewed preliminary versions of the manuscript and/or CD-ROM; their thoughtful feedback helped to shape what you see here. Thanks also to Allan Wylde, the editor of my previous book as well as this one, and Keisha Sherbecoe and Robert Wexler of Springer-Verlag. I learned much of what I know about algorithms along with my graduate students Yaw-Ling Lin, Sundaram Gopalakrishnan, Ting Chen, Francine Evans, Harald Rau, Ricky Bradley, and Dimitris Margaritis. They are the real heroes of many of the war stories related within. Much of the rest I have learned with my Stony Brook friends and colleagues Estie Arkin and Joe Mitchell, who have always been a pleasure to work and be with. Finally, I'd like to send personal thanks to several people. Mom, Dad, Len, and Rob all provided moral support. Michael Brochstein took charge of organizing my social life, thus freeing time for me to actually write the book. Through his good offices I met Renee. Her love and patience since then have made it all worthwhile.

Next: Caveat Up: The Algorithm Design Manual Previous: Preface Algorithms

file:///E|/BOOK/BOOK/NODE2.HTM (1 of 2) [19/1/2003 1:27:33]

Acknowledgments

Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE2.HTM (2 of 2) [19/1/2003 1:27:33]

Caveat

Next: Contents Up: The Algorithm Design Manual Previous: Acknowledgments

Caveat It is traditional for the author to magnanimously accept the blame for whatever deficiencies remain. I don't. Any errors, deficiencies, or problems in this book are somebody else's fault, but I would appreciate knowing about them so as to determine who is to blame. Steven S. Skiena Department of Computer Science State University of New York Stony Brook, NY 11794-4400 http://www.cs.sunysb.edu/ skiena May 1997

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE3.HTM [19/1/2003 1:27:33]

Contents

Next: Techniques Up: The Algorithm Design Manual Previous: Caveat

Contents ●

Techniques ❍ Introduction to Algorithms ■ Correctness and Efficiency ■ Correctness ■ Efficiency ■ Expressing Algorithms ■ Keeping Score ■ The RAM Model of Computation ■ Best, Worst, and Average-Case Complexity ■ The Big Oh Notation ■ Growth Rates ■ Logarithms ■ Modeling the Problem ■ About the War Stories ■ War Story: Psychic Modeling ■ Exercises ❍ Data Structures and Sorting ■ Fundamental Data Types ■ Containers ■ Dictionaries ■ Binary Search Trees ■ Priority Queues ■ Specialized Data Structures ■ Sorting ■ Applications of Sorting ■ Approaches to Sorting ■ Data Structures ■ Incremental Insertion ■ Divide and Conquer ■ Randomization ■ Bucketing Techniques ■ War Story: Stripping Triangulations

file:///E|/BOOK/BOOK/NODE4.HTM (1 of 7) [19/1/2003 1:27:35]

Contents

War Story: Mystery of the Pyramids ■ War Story: String 'em Up ■ Exercises Breaking Problems Down ■ Dynamic Programming ■ Fibonacci numbers ■ The Partition Problem ■ Approximate String Matching ■ Longest Increasing Sequence ■ Minimum Weight Triangulation ■ Limitations of Dynamic Programming ■ War Story: Evolution of the Lobster ■ War Story: What's Past is Prolog ■ War Story: Text Compression for Bar Codes ■ Divide and Conquer ■ Fast Exponentiation ■ Binary Search ■ Square and Other Roots ■ Exercises Graph Algorithms ■ The Friendship Graph ■ Data Structures for Graphs ■ War Story: Getting the Graph ■ Traversing a Graph ■ Breadth-First Search ■ Depth-First Search ■ Applications of Graph Traversal ■ Connected Components ■ Tree and Cycle Detection ■ Two-Coloring Graphs ■ Topological Sorting ■ Articulation Vertices ■ Modeling Graph Problems ■ Minimum Spanning Trees ■ Prim's Algorithm ■ Kruskal's Algorithm ■ Shortest Paths ■ Dijkstra's Algorithm ■

❍

❍

file:///E|/BOOK/BOOK/NODE4.HTM (2 of 7) [19/1/2003 1:27:35]

Contents

All-Pairs Shortest Path ■ War Story: Nothing but Nets ■ War Story: Dialing for Documents ■ Exercises Combinatorial Search and Heuristic Methods ■ Backtracking ■ Constructing All Subsets ■ Constructing All Permutations ■ Constructing All Paths in a Graph ■ Search Pruning ■ Bandwidth Minimization ■ War Story: Covering Chessboards ■ Heuristic Methods ■ Simulated Annealing ■ Traveling Salesman Problem ■ Maximum Cut ■ Independent Set ■ Circuit Board Placement ■ Neural Networks ■ Genetic Algorithms ■ War Story: Annealing Arrays ■ Parallel Algorithms ■ War Story: Going Nowhere Fast ■ Exercises Intractable Problems and Approximations ■ Problems and Reductions ■ Simple Reductions ■ Hamiltonian Cycles ■ Independent Set and Vertex Cover ■ Clique and Independent Set ■ Satisfiability ■ The Theory of NP-Completeness ■ 3-Satisfiability ■ Difficult Reductions ■ Integer Programming ■ Vertex Cover ■ Other NP-Complete Problems ■ The Art of Proving Hardness ■

❍

❍

file:///E|/BOOK/BOOK/NODE4.HTM (3 of 7) [19/1/2003 1:27:35]

Contents

War Story: Hard Against the Clock ■ Approximation Algorithms ■ Approximating Vertex Cover ■ The Euclidean Traveling Salesman ■ Exercises ❍ How to Design Algorithms Resources ❍ A Catalog of Algorithmic Problems ■ Data Structures ■ Dictionaries ■ Priority Queues ■ Suffix Trees and Arrays ■ Graph Data Structures ■ Set Data Structures ■ Kd-Trees ■ Numerical Problems ■ Solving Linear Equations ■ Bandwidth Reduction ■ Matrix Multiplication ■ Determinants and Permanents ■ Constrained and Unconstrained Optimization ■ Linear Programming ■ Random Number Generation ■ Factoring and Primality Testing ■ Arbitrary-Precision Arithmetic ■ Knapsack Problem ■ Discrete Fourier Transform ■ Combinatorial Problems ■ Sorting ■ Searching ■ Median and Selection ■ Generating Permutations ■ Generating Subsets ■ Generating Partitions ■ Generating Graphs ■ Calendrical Calculations ■ Job Scheduling ■ Satisfiability ■

●

file:///E|/BOOK/BOOK/NODE4.HTM (4 of 7) [19/1/2003 1:27:35]

Contents ■

■

■

Graph Problems: Polynomial-Time ■ Connected Components ■ Topological Sorting ■ Minimum Spanning Tree ■ Shortest Path ■ Transitive Closure and Reduction ■ Matching ■ Eulerian Cycle / Chinese Postman ■ Edge and Vertex Connectivity ■ Network Flow ■ Drawing Graphs Nicely ■ Drawing Trees ■ Planarity Detection and Embedding Graph Problems: Hard Problems ■ Clique ■ Independent Set ■ Vertex Cover ■ Traveling Salesman Problem ■ Hamiltonian Cycle ■ Graph Partition ■ Vertex Coloring ■ Edge Coloring ■ Graph Isomorphism ■ Steiner Tree ■ Feedback Edge/Vertex Set Computational Geometry ■ Robust Geometric Primitives ■ Convex Hull ■ Triangulation ■ Voronoi Diagrams ■ Nearest Neighbor Search ■ Range Search ■ Point Location ■ Intersection Detection ■ Bin Packing ■ Medial-Axis Transformation ■ Polygon Partitioning ■ Simplifying Polygons

file:///E|/BOOK/BOOK/NODE4.HTM (5 of 7) [19/1/2003 1:27:35]

Contents

Shape Similarity ■ Motion Planning ■ Maintaining Line Arrangements ■ Minkowski Sum ■ Set and String Problems ■ Set Cover ■ Set Packing ■ String Matching ■ Approximate String Matching ■ Text Compression ■ Cryptography ■ Finite State Machine Minimization ■ Longest Common Substring ■ Shortest Common Superstring ❍ Algorithmic Resources ■ Software systems ■ LEDA ■ Netlib ■ Collected Algorithms of the ACM ■ The Stanford GraphBase ■ Combinatorica ■ Algorithm Animations with XTango ■ Programs from Books ■ Discrete Optimization Algorithms in Pascal ■ Handbook of Data Structures and Algorithms ■ Combinatorial Algorithms for Computers and Calculators ■ Algorithms from P to NP ■ Computational Geometry in C ■ Algorithms in C++ ■ Data Sources ■ Textbooks ■ On-Line Resources ■ Literature ■ People ■ Software ■ Professional Consulting Services References Index ■

● ●

file:///E|/BOOK/BOOK/NODE4.HTM (6 of 7) [19/1/2003 1:27:35]

Contents ●

About this document ...

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE4.HTM (7 of 7) [19/1/2003 1:27:35]

Index

Next: About this document Up: The Algorithm Design Manual Previous: References

Index This index provides fast access to important keywords and topics in the on-line Book. The menu below partitions the index entries by the first letter of the alphabet, for ease of access. A full document index is also provided. Be aware that the index pointer typically resides at the end of the relevant paragraph in the document, so we recommend scrolling once towards the front of the document before reading. A similar index has been provided for the Lecture Notes, which may also be of interest.

A B C D E I

J

G

H

K L M N O

P

Q R S

T U

F

V W X

Y Z Complete Index (note: the complete index is large; it will take a bit of time to load)

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE233.HTM [19/1/2003 1:27:36]

The Algorithm Design Manual

The Algorithm Design Manual The CD-ROM Steven S. Skiena Department of Computer Science State University of New York Stony Brook, NY 11794-4400

What is a manual without software? The electronic supplement to this book is a ISO-9660 compatible, multiplatform CD-ROM, which can be viewed using Netscape, Microsoft Explorer, or any other WWW browser. This CD-ROM contains: ●

●

●

The Algorithm Design Manual: Hypertext Edition A complete hypertext version of the full printed book. Indeed, the extensive cross-references within the book are best followed using the hypertext version. The Algorithm Repository Website -- The source code and URLs for all cited implementations, mirroring the Stony Brook Algorithm Repository WWW site. Programs in C, C++, Fortran, and Pascal are included, providing an average of four different implementations for each algorithmic problem. Algorithms Lectures -- More than 30 hours of audio lectures on the design and analysis of algorithms are provided, all keyed to on-line lecture notes. Following these lectures provides another approach to learning algorithm design techniques. Listening to all the audio is analogous to taking a one-semester

file:///E|/INDEX.HTM (1 of 2) [19/1/2003 1:27:37]

The Algorithm Design Manual

●

● ● ● ●

● ●

college course on algorithms! Bibliographic References -- Pointers to the net's most important collections of references on algorithms. Local copies of two large bibliographies are included. About the Book Copyright Notice and Disclaimers Graphics Gallery Send us Mail Guide to Configuring Browsers Thanks!

file:///E|/INDEX.HTM (2 of 2) [19/1/2003 1:27:37]

Lecture Notes -- Analysis of Algorithms

CSE 373/548 - Analysis of Algorithms Lecture Notes with Audio Steven Skiena Department of Computer Science SUNY Stony Brook In Spring 1996, I taught my Analysis of Algorithms course via EngiNet, the SUNY Stony Brook distance learning program. Each of my lectures that semester was videotaped, and the tapes made available to off-site students. I found it an enjoyable experience. As an experiment in using the Internet for distance learning, we have digitized the complete audio of all 23 lectures, and have made this available on the WWW. We partitioned the full audio track into sound clips, each corresponding to one page of lecture notes, and linked them to the associated text. In a real sense, listening to all the audio is analogous to sitting through a one-semester college course on algorithms! Properly compressed, the full semester's audio requires less than 300 megabytes of storage, which is much less than I would have imagined. The entire semesters lectures, over thirty hours of audio files, fit comfortably on The Algorithm Design Manual CD-ROM, which also includes a hypertext version of the book and a substantial amount of software. All exercise numbers refer to Corman, Leiserson, and Rivest's Introduction to Algorithms, the textbook I used that particular year. The sound quality is amazingly good, considering it was me that they were taping. Unfortunately, the Shockwave format we used is only supported under Windows and Macintoshes, so the sound cannot be heard under UNIX. On certain browsers, a new window is opened for each sound bite, so be sure to close these windows before they cause trouble. Because of space requirements, we did not digitize much of the corresponding video, which would have made the presentation even more interesting. Still, I hope you find that these audio lectures expand your understanding of both algorithm design and educational multimedia. The full video tapes themselves are also available. ●

Postscript lecture transparencies file:///E|/LEC/LECTURES/ALL.HTM (1 of 3) [19/1/2003 1:27:38]

Lecture Notes -- Analysis of Algorithms ● ● ●

A Guide to Configuring Browsers Binary Search Animation Other Algorithms Courses

Links To Individual Lectures ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

Lecture 1 - analyzing algorithms Lecture 2 - asymptotic notation Lecture 3 - recurrence relations Lecture 4 - heapsort Lecture 5 - quicksort Lecture 6 - linear sorting Lecture 7 - elementary data structures Lecture 8 - binary trees Lecture 9 - catch up Lecture 10 - tree restructuring Lecture 11 - backtracking Lecture 12 - introduction to dynamic programming Lecture 13 - dynamic programming applications Lecture 14 - data structures for graphs Lecture 15 - DFS and BFS Lecture 16 - applications of DFS and BFS Lecture 17 - minimum spanning trees Lecture 18 - shortest path algorthms Lecture 19 - satisfiability Lecture 20 - integer programming Lecture 21 - vertex cover Lecture 22 - techniques for proving hardness Lecture 23 - approximation algorithms and Cook's theorem Index About this document ...

Next: Lecture 1 - analyzing Up: Main Page

file:///E|/LEC/LECTURES/ALL.HTM (2 of 3) [19/1/2003 1:27:38]

Lecture Notes -- Analysis of Algorithms

Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTURES/ALL.HTM (3 of 3) [19/1/2003 1:27:38]

The Stony Brook Algorithm Repository

The Stony Brook Algorithm Repository Steven S. Skiena Department of Computer Science State University of New York Stony Brook, NY 11794-4400 This WWW page is intended to serve as a comprehensive collection of algorithm implementations for over seventy of the most fundamental problems in combinatorial algorithms. The problem taxonomy, implementations, and supporting material are all drawn from my book The Algorithm Design Manual . Since the practical person is more often looking for a program than an algorithm, we provide pointers to solid implementations of useful algorithms, when they are available. Because of the volatility of the WWW, we provide local copies for many of the implementations. We encourage you to get them from the original sites instead of Stony Brook, because the version on the original site is more likely to be maintained. Further, there are often supporting files and documentation which we did not copy, and which may be of interest to you. The local copies of large implementations are maintained as gzip tar archives and, where available, DOS zip archives. Software for decoding these formats is readily available . Many of these codes have been made available for research or educational use, although commercial use requires a licensing arrangement with the author. Licensing terms from academic institutions are usually surprisingly modest. The recognition that industry is using a particular code is important to the authors, often more important than the money. This can lead to enhanced support or future releases of the software. Do the right thing and get a license -- information about terms or who to contact is usually available embedded within the documentation, or available at the original source site. Use at your own risk. The author and Springer-Verlag make no representations, express or implied, with respect to any software or documentation we describe. The author and Springer-Verlag shall in no event be liable for any indirect, incidental, or consequential damages.

file:///E|/WEBSITE/INDEX.HTM (1 of 3) [19/1/2003 1:27:40]

The Stony Brook Algorithm Repository

About the Book

Graphics Gallery

Graphic Index

Audio Algorithm Lectures

Bibliography References

Data Sources

About the Ratings

Copyright Notice and Disclaimers

Top-Level CD-ROM Page

Problems by Category ● ● ● ● ● ● ●

1.1 Data Structures 1.2 Numerical Problems 1.3 Combinatorial Problems 1.4 Graph Problems -- polynomial-time problems 1.5 Graph Problems -- hard problems 1.6 Computational Geometry 1.7 Set and String Problems

Implementations By Language ● ● ● ● ● ●

C++ Language Implementations C Language Implementations Pascal Language Implementations FORTRAN Language Implementations Mathematica Language Implementations Lisp Language Implementations

Other information on this site ● ● ● ● ● ● ●

Order the Book Most Wanted List Tools and Utilities Repository Citations Video tapes Feedback Thanks! file:///E|/WEBSITE/INDEX.HTM (2 of 3) [19/1/2003 1:27:40]

The Stony Brook Algorithm Repository

Send us mail Web-site regenerated on Tue Jun 03, 1997

file:///E|/WEBSITE/INDEX.HTM (3 of 3) [19/1/2003 1:27:40]

Techniques

Next: Introduction to Algorithms Up: The Algorithm Design Manual Previous: Contents

Techniques

●

●

●

Introduction to Algorithms ❍ Correctness and Efficiency ❍ Expressing Algorithms ❍ Keeping Score ❍ The Big Oh Notation ❍ Growth Rates ❍ Logarithms ❍ Modeling the Problem ❍ About the War Stories ❍ War Story: Psychic Modeling ❍ Exercises Data Structures and Sorting ❍ Fundamental Data Types ❍ Specialized Data Structures ❍ Sorting ❍ Applications of Sorting ❍ Approaches to Sorting ❍ War Story: Stripping Triangulations ❍ War Story: Mystery of the Pyramids ❍ War Story: String 'em Up ❍ Exercises Breaking Problems Down ❍ Dynamic Programming ❍ Limitations of Dynamic Programming ❍ War Story: Evolution of the Lobster ❍ War Story: What's Past is Prolog ❍ War Story: Text Compression for Bar Codes

file:///E|/BOOK/BOOK/NODE5.HTM (1 of 3) [19/1/2003 1:27:41]

Techniques

Divide and Conquer ❍ Exercises Graph Algorithms ❍ The Friendship Graph ❍ Data Structures for Graphs ❍ War Story: Getting the Graph ❍ Traversing a Graph ❍ Applications of Graph Traversal ❍ Modeling Graph Problems ❍ Minimum Spanning Trees ❍ Shortest Paths ❍ War Story: Nothing but Nets ❍ War Story: Dialing for Documents ❍ Exercises Combinatorial Search and Heuristic Methods ❍ Backtracking ❍ Search Pruning ❍ Bandwidth Minimization ❍ War Story: Covering Chessboards ❍ Heuristic Methods ❍ War Story: Annealing Arrays ❍ Parallel Algorithms ❍ War Story: Going Nowhere Fast ❍ Exercises Intractable Problems and Approximations ❍ Problems and Reductions ❍ Simple Reductions ❍ Satisfiability ❍ Difficult Reductions ❍ Other NP-Complete Problems ❍ The Art of Proving Hardness ❍ War Story: Hard Against the Clock ❍ Approximation Algorithms ❍ Exercises How to Design Algorithms ❍

●

●

●

●

file:///E|/BOOK/BOOK/NODE5.HTM (2 of 3) [19/1/2003 1:27:41]

Techniques

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE5.HTM (3 of 3) [19/1/2003 1:27:41]

Introduction to Algorithms

Next: Correctness and Efficiency Up: Techniques Previous: Techniques

Introduction to Algorithms What is an algorithm? An algorithm is a procedure to accomplish a specific task. It is the idea behind any computer program. To be interesting, an algorithm has to solve a general, well-specified problem. An algorithmic problem is specified by describing the complete set of instances it must work on and what properties the output must have as a result of running on one of these instances. This distinction between a problem and an instance of a problem is fundamental. For example, the algorithmic problem known as sorting is defined as follows: Input: A sequence of n keys

.

Output: The permutation (reordering) of the input sequence such that . An instance of sorting might be an array of names, such as {Mike, Bob, Sally, Jill, Jan}, or a list of numbers like {154, 245, 568, 324, 654, 324}. Determining whether you in fact have a general problem to deal with, as opposed to an instance of a problem, is your first step towards solving it. This is true in algorithms as it is in life. An algorithm is a procedure that takes any of the possible input instances and transforms it to the desired output. There are many different algorithms for solving the problem of sorting. For example, one method for sorting starts with a single element (thus forming a trivially sorted list) and then incrementally inserts the remaining elements so that the list stays sorted. This algorithm, insertion sort, is described below:

InsertionSort(A)

for i = 1 to n-1 do

for j = i+1 to 2 do

if (A[j] < A[j-1]) then swap(A[j],A[j1]) Note the generality of this algorithm. It works equally well on names as it does on numbers, given the appropriate < comparison operation to test which of the two keys should appear first in sorted order. Given our definition of the sorting problem, it can be readily verified that this algorithm correctly orders every possible input instance. In this chapter, we introduce the desirable properties that good algorithms have, as well as how to measure whether a given algorithm achieves these goals. Assessing algorithmic performance requires a modest amount of mathematical notation, which we also present. Although initially intimidating, this notation proves essential for us to compare algorithms and design more efficient ones. While the hopelessly ``practical'' person may blanch at the notion of theoretical analysis, we present this material because it is useful. In particular, this chapter offers the following ``take-home'' lessons:

file:///E|/BOOK/BOOK/NODE6.HTM (1 of 2) [19/1/2003 1:27:42]

Introduction to Algorithms ●

● ●

● ●

●

● ●

● ● ● ● ● ● ●

Reasonable-looking algorithms can easily be incorrect. Algorithm correctness is a property that must be carefully demonstrated. Algorithms can be understood and studied in a machine independent way. The ``big Oh'' notation and worst-case analysis are tools that greatly simplify our ability to compare the efficiency of algorithms. We seek algorithms whose running times grow logarithmically, because grows very slowly with increasing n. Modeling your application in terms of well-defined structures and algorithms is the most important single step towards a solution.

Correctness and Efficiency ❍ Correctness ❍ Efficiency Expressing Algorithms Keeping Score ❍ The RAM Model of Computation ❍ Best, Worst, and Average-Case Complexity The Big Oh Notation Growth Rates Logarithms Modeling the Problem About the War Stories War Story: Psychic Modeling Exercises

Next: Correctness and Efficiency Up: Techniques Previous: Techniques Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE6.HTM (2 of 2) [19/1/2003 1:27:42]

Data Structures and Sorting

Next: Fundamental Data Types Up: Techniques Previous: Implementation Challenges

Data Structures and Sorting When things go right, changing a data structure in a slow program works the same way an organ transplant does in a sick patient. For several classes of abstract data types, such as containers, dictionaries, and priority queues, there exist many different but functionally equivalent data structures that implement the given data type. Changing the data structure does not change the correctness of the program, since we presumably replace a correct implementation with a different correct implementation. However, because the new implementation of the data type realizes different tradeoffs in the time to execute various operations, the total performance of an application can improve dramatically. Like a patient in need of a transplant, only one part might need to be replaced in order to fix the problem. It is obviously better to be born with a good heart than have to wait for a replacement. Similarly, the maximum benefit from good data structures results from designing your program around them in the first place. Still, it is important to build your programs so that alternative implementations can be tried. This involves separating the internals of the data structure (be it a tree, a hash table, or a sorted array) from its interface (operations like search, insert, delete). Such data abstraction is an important part of producing clean, readable, and modifiable programs. We will not dwell on such software engineering issues here, but such a design is critical if you are to experiment with the impact of different implementations on performance. In this chapter we will also discuss sorting, stressing how sorting can be applied to solve other problems more than the details of specific sorting algorithms. In this sense, sorting behaves more like a data structure than a problem in its own right. Sorting is also represented by a significant entry in the problem catalog; namely Section

.

The key take-home lessons of this chapter are: ●

●

●

●

Building algorithms around data structures such as dictionaries and priority queues leads to both clean structure and good performance. Picking the wrong data structure for the job can be disastrous in terms of performance. Picking the very best data structure is often not as critical, for there are typically several choices that perform similarly. Sorting lies at the heart of many different algorithms. Sorting the data is one of the first things any algorithm designer should try in the quest for efficiency. Sorting can be used to illustrate most algorithm design paradigms. Data structure techniques,

file:///E|/BOOK/BOOK/NODE22.HTM (1 of 2) [19/1/2003 1:27:43]

Data Structures and Sorting

divide-and-conquer, randomization, and incremental construction all lead to popular sorting algorithms.

●

● ● ● ●

● ● ● ●

Fundamental Data Types ❍ Containers ❍ Dictionaries ❍ Binary Search Trees ❍ Priority Queues Specialized Data Structures Sorting Applications of Sorting Approaches to Sorting ❍ Data Structures ❍ Incremental Insertion ❍ Divide and Conquer ❍ Randomization ❍ Bucketing Techniques War Story: Stripping Triangulations War Story: Mystery of the Pyramids War Story: String 'em Up Exercises

Next: Fundamental Data Types Up: Techniques Previous: Implementation Challenges Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE22.HTM (2 of 2) [19/1/2003 1:27:43]

Breaking Problems Down

Next: Dynamic Programming Up: Techniques Previous: Implementation Challenges

Breaking Problems Down One of the most powerful techniques for solving problems is to break them down into smaller, more easily solved pieces. Smaller problems are less overwhelming, and they permit us to focus on details that are lost when we are studying the entire problem. For example, whenever we can break the problem into smaller instances of the same type of problem, a recursive algorithm starts to become apparent. Two important algorithm design paradigms are based on breaking problems down into smaller problems. Dynamic programming typically removes one element from the problem, solves the smaller problem, and then uses the solution to this smaller problem to add back the element in the proper way. Divide and conquer typically splits the problem in half, solves each half, then stitches the halves back together to form a full solution. Both of these techniques are important to know about. Dynamic programming in particular is a misunderstood and underappreciated technique. To demonstrate its utility in practice, we present no fewer than three war stories where dynamic programming played the decisive role. The take-home lessons for this chapter include: ●

●

●

●

●

Many objects have an inherent left-to-right ordering among their elements, such as characters in a string, elements of a permutation, points around a polygon, or leaves in a search tree. For any optimization problem on such left-to-right objects, dynamic programming will likely lead to an efficient algorithm to find the best solution. Without an inherent left-to-right ordering on the objects, dynamic programming is usually doomed to require exponential space and time. Once you understand dynamic programming, it can be easier to work out such algorithms from scratch than to try to look them up. The global optimum (found, for example, using dynamic programming) is often noticeably better than the solution found by typical heuristics. How important this improvement is depends upon your application, but it can never hurt. Binary search and its variants are the quintessential divide-and-conquer algorithms.

file:///E|/BOOK/BOOK/NODE42.HTM (1 of 2) [19/1/2003 1:27:43]

Breaking Problems Down ●

● ● ● ● ●

●

Dynamic Programming ❍ Fibonacci numbers ❍ The Partition Problem ❍ Approximate String Matching ❍ Longest Increasing Sequence ❍ Minimum Weight Triangulation Limitations of Dynamic Programming War Story: Evolution of the Lobster War Story: What's Past is Prolog War Story: Text Compression for Bar Codes Divide and Conquer ❍ Fast Exponentiation ❍ Binary Search ❍ Square and Other Roots Exercises

Next: Dynamic Programming Up: Techniques Previous: Implementation Challenges Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE42.HTM (2 of 2) [19/1/2003 1:27:43]

Graph Algorithms

Next: The Friendship Graph Up: Techniques Previous: Implementation Challenges

Graph Algorithms A graph G=(V,E) consists of a set of vertices V together with a set E of vertex pairs or edges. Graphs are important because they can be used to represent essentially any relationship. For example, graphs can model a network of roads, with cities as vertices and roads between cities as edges, as shown in Figure . Electronic circuits can also be modeled as graphs, with junctions as vertices and components as edges.

Figure: Modeling road networks and electronic circuits as graphs The key to understanding many algorithmic problems is to think of them in terms of graphs. Graph theory provides a language for talking about the properties of graphs, and it is amazing how often messy applied problems have a simple description and solution in terms of classical graph properties. Designing truly novel graph algorithms is a very difficult task. The key to using graph algorithms effectively in applications lies in correctly modeling your problem as a standard graph property, so you can take advantage of existing algorithms. Becoming familiar with many different graph algorithmic problems is more important than understanding the details of particular graph algorithms, particularly since Part II of this book can point you to an implementation as soon as you know the name of your problem. In this chapter, we will present basic data structures and traversal operations for graphs, which will enable you to cobble together solutions to rudimentary graph problems. We will also describe more sophisticated algorithms for problems like shortest paths and minimum spanning trees in some detail. But we stress the primary importance of correctly modeling your problem. Time spent browsing through the catalog now will leave you better informed of your options when a real job arises. file:///E|/BOOK/BOOK2/NODE59.HTM (1 of 3) [19/1/2003 1:27:44]

Graph Algorithms

The take-home lessons of this chapter include: ● ●

●

● ● ● ●

●

● ●

●

● ● ●

Graphs can be used to model a wide variety of structures and relationships. Properly formulated, most applications of graphs can be reduced to standard graph properties and using well-known algorithms. These include minimum spanning trees, shortest paths, and several problems presented in the catalog. Breadth-first and depth-first search provide mechanisms to visit each edge and vertex of the graph. They prove the basis of most simple, efficient graph algorithms.

The Friendship Graph Data Structures for Graphs War Story: Getting the Graph Traversing a Graph ❍ Breadth-First Search ❍ Depth-First Search Applications of Graph Traversal ❍ Connected Components ❍ Tree and Cycle Detection ❍ Two-Coloring Graphs ❍ Topological Sorting ❍ Articulation Vertices Modeling Graph Problems Minimum Spanning Trees ❍ Prim's Algorithm ❍ Kruskal's Algorithm Shortest Paths ❍ Dijkstra's Algorithm ❍ All-Pairs Shortest Path War Story: Nothing but Nets War Story: Dialing for Documents Exercises

file:///E|/BOOK/BOOK2/NODE59.HTM (2 of 3) [19/1/2003 1:27:44]

Graph Algorithms

Next: The Friendship Graph Up: Techniques Previous: Implementation Challenges Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE59.HTM (3 of 3) [19/1/2003 1:27:44]

Combinatorial Search and Heuristic Methods

Next: Backtracking Up: Techniques Previous: Implementation Challenges

Combinatorial Search and Heuristic Methods We have seen how clever algorithms can reduce the complexity of sorting from to , which is good. However, the algorithmic stakes can be even higher for combinatorially explosive problems, whose time will make clear the grows exponentially in the size of the problem. Looking back at Figure limitations of exponential-time algorithms on even modest-sized problems. By using exhaustive search techniques, we can solve small problems to optimality, although the time complexity may be enormous. For certain applications, it may well pay to spend extra time to be certain of the optimal solution. A good example occurs in testing a circuit or a program on all possible inputs. You can prove the correctness of the device by trying all possible inputs and verifying that they give the correct answer. Proving such correctness is a property to be proud of. However, claiming that it works correctly on all the inputs you tried is worth much, much less. In this section, we present backtracking as a technique for listing all configurations representing possible solutions for a combinatorial algorithm problem. We then discuss techniques for pruning search that significantly improve efficiency by eliminating irrelevant configurations from consideration. We illustrate the power of clever pruning techniques to speed up real search applications. For problems that are too large to contemplate using brute-force combinatorial search, we introduce heuristic methods such as simulated annealing. Such heuristic methods are an important weapon in the practical algorist's arsenal. The take-home lessons from this chapter are: ●

●

●

Combinatorial search, augmented with tree pruning techniques, can be used to find the optimal solution of small optimization problems. How small depends upon the specific problem, but the size limit is likely to be somewhere between items. Clever pruning techniques can speed up combinatorial search to an amazing extent. Proper pruning will have a greater impact on search time than any other factor. Simulated annealing is a simple but effective technique to efficiently obtain good but not optimal solutions to combinatorial search problems.

file:///E|/BOOK/BOOK2/NODE83.HTM (1 of 2) [19/1/2003 1:27:45]

Combinatorial Search and Heuristic Methods

●

● ● ● ●

● ● ● ●

Backtracking ❍ Constructing All Subsets ❍ Constructing All Permutations ❍ Constructing All Paths in a Graph Search Pruning Bandwidth Minimization War Story: Covering Chessboards Heuristic Methods ❍ Simulated Annealing ❍ Neural Networks ❍ Genetic Algorithms War Story: Annealing Arrays Parallel Algorithms War Story: Going Nowhere Fast Exercises

Next: Backtracking Up: Techniques Previous: Implementation Challenges Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE83.HTM (2 of 2) [19/1/2003 1:27:45]

Intractable Problems and Approximations

Next: Problems and Reductions Up: Techniques Previous: Implementation Challenges

Intractable Problems and Approximations In this chapter, we will concentrate on techniques for proving that no efficient algorithm exists for a given problem. The practical reader is probably squirming at the notion of proving anything and will be particularly alarmed at the idea of investing time to prove that something does not exist. Why will you be better off knowing that something you don't know how to do in fact can't be done at all? The truth is that the theory of NP-completeness is an immensely useful tool for the algorithm designer, even though all it does is provide negative results. That noted algorithm designer Sherlock Holmes once said, ``When you have eliminated the impossible, what remains, however improbable, must be the truth.'' The theory of NP-completeness enables the algorithm designer to focus her efforts more productively, by revealing that the search for an efficient algorithm for this particular problem is doomed to failure. When one fails to show that a problem is hard, that means there is likely an algorithm that solves it efficiently. Two of the war stories in this book describe happy results springing from bogus claims of hardness. The theory of NP-completeness also enables us to identify exactly what properties make a particular problem hard, thus providing direction for us to model it in different ways or exploit more benevolent characteristics of the problem. Developing a sense for which problems are hard and which are not is a fundamental skill for algorithm designers, and it can come only from hands-on experience proving hardness. We will not discuss the complexity-theoretic aspects of NP-completeness in depth, limiting our treatment to the fundamental concept of reductions, which show the equivalence of pairs of problems. For a discussion, we refer the reader to [GJ79], the truly essential reference on the theory of intractability. The take-home lessons from this chapter are: ●

●

Reductions are a way to show that two problems are essentially identical. A fast algorithm for one of the problems implies a fast algorithm for the other. In practice, a small set of NP-complete problems (3-SAT, vertex cover, integer partition, and Hamiltonian cycle) suffice to prove the hardness of most other hard problems.

file:///E|/BOOK/BOOK3/NODE104.HTM (1 of 2) [19/1/2003 1:27:46]

Intractable Problems and Approximations ●

● ●

●

●

● ● ● ●

●

Approximation algorithms guarantee answers that are always close to the optimal solution and can provide an approach to dealing with NP-complete problems.

Problems and Reductions Simple Reductions ❍ Hamiltonian Cycles ❍ Independent Set and Vertex Cover ❍ Clique and Independent Set Satisfiability ❍ The Theory of NP-Completeness ❍ 3-Satisfiability Difficult Reductions ❍ Integer Programming ❍ Vertex Cover Other NP-Complete Problems The Art of Proving Hardness War Story: Hard Against the Clock Approximation Algorithms ❍ Approximating Vertex Cover ❍ The Euclidean Traveling Salesman Exercises

Next: Problems and Reductions Up: Techniques Previous: Implementation Challenges Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE104.HTM (2 of 2) [19/1/2003 1:27:46]

How to Design Algorithms

Next: Resources Up: Techniques Previous: Implementation Challenges

How to Design Algorithms Designing the right algorithm for a given application is a difficult job. It requires a major creative act, taking a problem and pulling a solution out of the ether. This is much more difficult than taking someone else's idea and modifying it or tweaking it to make it a little better. The space of choices you can make in algorithm design is enormous, enough to leave you plenty of freedom to hang yourself. This book is designed to make you a better algorithm designer. The techniques presented in Part I of this book provide the basic ideas underlying all combinatorial algorithms. The problem catalog of Part II will help you with modeling your application and point you in the right direction of an algorithm or implementation. However, being a successful algorithm designer requires more than book knowledge; it requires a certain attitude, the right problem-solving approach. It is difficult to teach this mindset in a book; yet getting it is essential to become a successful designer. The key to algorithm design (or any other problem-solving task) is to proceed by asking yourself a sequence of questions to guide your thought process. What if we do this? What if we do that? Should you get stuck on the problem, the best thing to do is move onto the next question. In any group brainstorming session, the most useful person in the room is the one who keeps asking, ``Why can't we do it this way?'' not the person who later tells them why. Because eventually she will stumble on an approach that can't be shot down. Towards this end, we provide below a sequence of questions to guide your search for the right algorithm for your problem. To use it effectively, you must not only ask the questions, but answer them. The key is working through the answers carefully, by writing them down in a log. The correct answer to, ``Can I do it this way?'' is never ``no,'' but ``no, because ....'' By clearly articulating your reasoning as to why something doesn't work, you can check if it really holds up or whether you have just glossed over a possibility that you didn't want to think hard enough about. You will be surprised how often the reason you can't find a convincing explanation for something is because your conclusion is wrong. An important distinction to keep aware of during any design process is the difference between strategy and tactics. Strategy represents the quest for the big picture, the framework around which we construct our path to the goal. Tactics are used to win the minor battles we must fight along the way. In problem solving, it is important to check repeatedly whether you are thinking on the right level. If you do not have a global strategy of how you are going to attack your problem, it is pointless to worry about the tactics. An example of a strategic question is, ``How best can I model my application as a graph algorithm

file:///E|/BOOK/BOOK3/NODE124.HTM (1 of 5) [19/1/2003 1:27:48]

How to Design Algorithms

problem?'' A tactical question might be, ``Should I use an adjacency list or adjacency matrix data structure to represent my graph?'' Of course, such tactical decisions are critical to the ultimate quality of the solution, but they can be properly evaluated only in light of a successful strategy. When faced with a design problem, too many people freeze up in their thinking. After reading or hearing the problem, they sit down and realize that they don't know what to do next. They stare into space, then panic, and finally end up settling for the first thing that comes to mind. Avoid this fate. Follow the sequence of questions provided below and in most of the catalog problem sections. We'll tell you what to do next! Obviously, the more experience you have with algorithm design techniques such as dynamic programming, graph algorithms, intractability, and data structures, the more successful you will be at working through the list of questions. Part I of this book has been designed to strengthen this technical background. However, it pays to work through these questions regardless of how strong your technical skills are. The earliest and most important questions on the list focus on obtaining a detailed understanding of the problem and do not require specific expertise. This list of questions was inspired by a passage in that wonderful book about the space program The Right Stuff [Wol79]. It concerned the radio transmissions from test pilots just before their planes crashed. One might have expected that they would panic, so that ground control would hear the pilot yelling Ahhhhhhhhhhh --, terminated only by the sound of smacking into a mountain. Instead, the pilots ran through a list of what their possible actions could be. I've tried the flaps. I've checked the engine. Still got two wings. I've reset the --. They had ``the Right Stuff.'' Because of this, they sometimes managed to miss the mountain. I hope this book and list will provide you with ``the Right Stuff'' to be an algorithm designer. And I hope it prevents you from smacking into any mountains along the way. 1. Do I really understand the problem? 1. What exactly does the input consist of? 2. What exactly are the desired results or output? 3. Can I construct an example input small enough to solve by hand? What happens when I try to solve it? 4. How important is it to my application that I always find an exact, optimal answer? Can I settle for something that is usually pretty good? 5. How large will a typical instance of my problem be? Will I be working on 10 items? 1,000 items? 1,000,000 items? 6. How important is speed in my application? Must the problem be solved within one second? One minute? One hour? One day? 7. How much time and effort can I invest in implementing my algorithm? Will I be limited to simple algorithms that can be coded up in a day, or do I have the freedom to experiment file:///E|/BOOK/BOOK3/NODE124.HTM (2 of 5) [19/1/2003 1:27:48]

How to Design Algorithms

with a couple of approaches and see which is best? 8. Am I trying to solve a numerical problem? A graph algorithm problem? A geometric problem? A string problem? A set problem? Might my problem be formulated in more than one way? Which formulation seems easiest? 2. Can I find a simple algorithm or heuristic for the problem? 1. Can I find an algorithm to solve my problem correctly by searching through all subsets or arrangements and picking the best one? 1. If so, why am I sure that this algorithm always gives the correct answer? 2. How do I measure the quality of a solution once I construct it? 3. Does this simple, slow solution run in polynomial or exponential time? Is my problem small enough that this brute-force solution will suffice? 4. If I can't find a slow, guaranteed correct algorithm, why am I certain that my problem is sufficiently well-defined to have a correct solution? 2. Can I solve my problem by repeatedly trying some simple rule, like picking the biggest item first? The smallest item first? A random item first? 1. If so, on what types of inputs does this heuristic work well? Do these correspond to the data that might arise in my application? 2. On what types of inputs does this heuristic work badly? If no such examples can be found, can I show that it always works well? 3. How fast does my heuristic come up with an answer? Does it have a simple implementation? 3. Is my problem in the catalog of algorithmic problems in the back of this book? 1. If it is, what is known about the problem? Is there an implementation available that I can use? 2. If I don't see my problem, did I look in the right place? Did I browse through all the pictures? Did I look in the index under all possible keywords? 3. Are there relevant resources available on the World-Wide Web? Did I do a Lycos, Alta Vista, or Yahoo search? Did I go to the WWW page associated with this book, ? 4. Are there special cases of the problem that I know how to solve exactly? 1. Can I solve the problem efficiently when I ignore some of the input parameters? 2. What happens when I set some of the input parameters to trivial values, such as 0 or 1? Does the problem become easier to solve? 3. Can I simplify the problem to the point where I can solve it efficiently? Is the problem now trivial or still interesting? file:///E|/BOOK/BOOK3/NODE124.HTM (3 of 5) [19/1/2003 1:27:48]

How to Design Algorithms

4. Once I know how to solve a certain special case, why can't this be generalized to a wider class of inputs? 5. Is my problem a special case of a more general problem in the catalog? 5. Which of the standard algorithm design paradigms are most relevant to my problem? 1. Is there a set of items that can be sorted by size or some key? Does this sorted order make it easier to find the answer? 2. Is there a way to split the problem in two smaller problems, perhaps by doing a binary search? How about partitioning the elements into big and small, or left and right? Does this suggest a divide-and-conquer algorithm? 3. Do the input objects or desired solution have a natural left-to-right order, such as characters in a string, elements of a permutation, or the leaves of a tree? If so, can I use dynamic programming to exploit this order? 4. Are there certain operations being repeatedly done on the same data, such as searching it for some element, or finding the largest/smallest remaining element? If so, can I use a data structure to speed up these queries? What about a dictionary/hash table or a heap/priority queue? 5. Can I use random sampling to select which object to pick next? What about constructing many random configurations and picking the best one? Can I use some kind of directed randomness like simulated annealing in order to zoom in on the best solution? 6. Can I formulate my problem as a linear program? How about an integer program? 7. Does my problem seem something like satisfiability, the traveling salesman problem, or some other NP-complete problem? If so, might the problem be NP-complete and thus not have an efficient algorithm? Is it in the problem list in the back of Garey and Johnson [GJ79]? 6. Am I still stumped? 1. Am I willing to spend money to hire an expert to tell me what to do? If so, check out the professional consulting services mentioned in Section . 2. Why don't I go back to the beginning and work through these questions again? Did any of my answers change during my latest trip through the list? Problem solving is not a science, but part art and part skill. It is one of the skills most worth developing. My favorite book on problem solving remains Pólya's How to Solve It [Pol57], which features a catalog of problem solving techniques that are fascinating to browse through, both before and after you have a problem.

file:///E|/BOOK/BOOK3/NODE124.HTM (4 of 5) [19/1/2003 1:27:48]

How to Design Algorithms

Next: Resources Up: Techniques Previous: Implementation Challenges Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE124.HTM (5 of 5) [19/1/2003 1:27:48]

Resources

Next: A Catalog of Algorithmic Up: The Algorithm Design Manual Previous: How to Design Algorithms

Resources

●

●

A Catalog of Algorithmic Problems ❍ Data Structures ❍ Numerical Problems ❍ Combinatorial Problems ❍ Graph Problems: Polynomial-Time ❍ Graph Problems: Hard Problems ❍ Computational Geometry ❍ Set and String Problems Algorithmic Resources ❍ Software systems ❍ Data Sources ❍ Textbooks ❍ On-Line Resources ❍ Professional Consulting Services

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE125.HTM [19/1/2003 1:27:48]

A Catalog of Algorithmic Problems

Next: Caveats Up: Resources Previous: Resources

A Catalog of Algorithmic Problems This is a catalog of algorithmic problems that arise commonly in practice. It describes what is known about them and gives suggestions about how best to proceed if the problem arises in your application. What is the best way to use this catalog? First, think a little about your problem. If you recall the name of your problem, look up the catalog entry in the index or table of contents and start reading. Read through the entire entry, since it contains pointers to other relevant problems that might be yours. If you don't find what you are looking for, leaf through the catalog, looking at the pictures and problem names to see if something strikes a chord. Don't be afraid to use the index, for every problem in the book is listed there under several possible keywords and applications. If you still don't find something relevant, your problem is either not suitable for attack by combinatorial algorithms or else you don't fully understand it. In either case, go back to step one. The catalog entries contain a variety of different types of information that have never been collected in one place before. Different fields in each entry present information of practical and historical interest. To make this catalog more easily accessible, we introduce each problem with a pair of graphics representing the problem instance or input on the left and the result of solving the problem on this instance on the right. We have invested considerable thought in selecting stylized images and examples that illustrate desired behaviors, more than just definitions. For example, the minimum spanning tree example illustrates how points can be clustered using minimum spanning trees. We hope that people without a handle on algorithmic terminology can flip through the pictures and identify which problems might be relevant to them. We augment these pictures with more formal written input and problem descriptions in order to eliminate any ambiguity inherent in a purely pictorial representation. Once you have identified your problem of interest, the discussion section tells you what you should do about it. We describe applications where the problem is likely to arise and special issues associated with data from them. We discuss the kind of results you can hope for or expect and, most importantly, what you should do to get them. For each problem, we outline a quick-and-dirty algorithm and pointers to algorithms to try next if the first attempt is not sufficient. We also identify other, related problems in the catalog.

file:///E|/BOOK/BOOK3/NODE126.HTM (1 of 4) [19/1/2003 1:27:49]

A Catalog of Algorithmic Problems

For most if not all of the problems presented, we identify readily available software implementations, which are discussed in the implementation field of each entry. Many of these routines are quite good, and they can perhaps be plugged directly into your application. Others will be incomplete or inadequate for production use, but they hopefully can provide a good model for your own implementation. In general, the implementations are listed in order of descending usefulness, but we will explicitly recommend the best one available for each problem if a clear winner exists. More detailed information . Essentially all of the implementations are for many of these implementations appears in Chapter available via the WWW site associated with this book, reachable at http://www.cs.sunysb.edu/ algorith. Finally, in deliberately smaller print, we discuss the history of each problem and present results of primarily theoretical interest. We have attempted to report the best results known for each problem and point out empirical comparisons of algorithms if they exist. This should be of interest to students and researchers, and also to practitioners for whom our recommended solutions prove inadequate and who need to know if anything better is possible.

●

●

●

Data Structures ❍ Dictionaries ❍ Priority Queues ❍ Suffix Trees and Arrays ❍ Graph Data Structures ❍ Set Data Structures ❍ Kd-Trees Numerical Problems ❍ Solving Linear Equations ❍ Bandwidth Reduction ❍ Matrix Multiplication ❍ Determinants and Permanents ❍ Constrained and Unconstrained Optimization ❍ Linear Programming ❍ Random Number Generation ❍ Factoring and Primality Testing ❍ Arbitrary-Precision Arithmetic ❍ Knapsack Problem ❍ Discrete Fourier Transform Combinatorial Problems

file:///E|/BOOK/BOOK3/NODE126.HTM (2 of 4) [19/1/2003 1:27:49]

A Catalog of Algorithmic Problems

Sorting ❍ Searching ❍ Median and Selection ❍ Generating Permutations ❍ Generating Subsets ❍ Generating Partitions ❍ Generating Graphs ❍ Calendrical Calculations ❍ Job Scheduling ❍ Satisfiability Graph Problems: Polynomial-Time ❍ Connected Components ❍ Topological Sorting ❍ Minimum Spanning Tree ❍ Shortest Path ❍ Transitive Closure and Reduction ❍ Matching ❍ Eulerian Cycle / Chinese Postman ❍ Edge and Vertex Connectivity ❍ Network Flow ❍ Drawing Graphs Nicely ❍ Drawing Trees ❍ Planarity Detection and Embedding Graph Problems: Hard Problems ❍ Clique ❍ Independent Set ❍ Vertex Cover ❍ Traveling Salesman Problem ❍ Hamiltonian Cycle ❍ Graph Partition ❍ Vertex Coloring ❍ Edge Coloring ❍ Graph Isomorphism ❍ Steiner Tree ❍ Feedback Edge/Vertex Set Computational Geometry ❍ Robust Geometric Primitives ❍ Convex Hull ❍

●

●

●

file:///E|/BOOK/BOOK3/NODE126.HTM (3 of 4) [19/1/2003 1:27:49]

A Catalog of Algorithmic Problems

Triangulation ❍ Voronoi Diagrams ❍ Nearest Neighbor Search ❍ Range Search ❍ Point Location ❍ Intersection Detection ❍ Bin Packing ❍ Medial-Axis Transformation ❍ Polygon Partitioning ❍ Simplifying Polygons ❍ Shape Similarity ❍ Motion Planning ❍ Maintaining Line Arrangements ❍ Minkowski Sum Set and String Problems ❍ Set Cover ❍ Set Packing ❍ String Matching ❍ Approximate String Matching ❍ Text Compression ❍ Cryptography ❍ Finite State Machine Minimization ❍ Longest Common Substring ❍ Shortest Common Superstring ❍

●

Next: Caveats Up: Resources Previous: Resources Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE126.HTM (4 of 4) [19/1/2003 1:27:49]

Algorithmic Resources

Next: Software systems Up: Resources Previous: Shortest Common Superstring

Algorithmic Resources This chapter describes resources that the practical algorithm designer should be familiar with. Although some of this information has appeared at various points in the catalog, the most important pointers have been collected here for general reference. These resources take the form of software systems and libraries, books, and other bibliographic sources. Many of the most interesting resources are available online.

●

● ● ●

●

Software systems ❍ LEDA ❍ Netlib ❍ The Stanford GraphBase ❍ Combinatorica ❍ Algorithm Animations with XTango ❍ Programs from Books Data Sources Textbooks On-Line Resources ❍ Literature ❍ People ❍ Software Professional Consulting Services

Algorithms Mon Jun 2 23:33:50 EDT 1997 file:///E|/BOOK/BOOK5/NODE210.HTM [19/1/2003 1:27:50]

References

Next: Index Up: The Algorithm Design Manual Previous: Professional Consulting Services

References AB95 D. Avis and D. Bremner. How good are convex hull algorithms? In Proc. 11th Annual ACM Symposium on Computational Geometry, pages 20-28, 1995. ABCC95 D. Applegate, R. Bixby, V. Chvatal, and W. Cook. Finding cuts in the TSP (a preliminary report). Technical Report 95-05, DIMACS, Rutgers University, Piscataway NJ, 1995. Abd80 N. N. Abdelmalek. A Fortran subroutine for the solution of overdetermined systems of linear equations. ACM Trans. Math. Softw., 6(2):228-230, June 1980. AC75 A. Aho and M. Corasick. Efficient string matching: an aid to bibliographic search. Communications of the ACM, 18:333-340, 1975. AC91 D. Applegate and W. Cook. A computational study of the job-shop scheduling problem. ORSA Journal on Computing, 3:149-156, 1991. ACH 91 E. M. Arkin, L. P. Chew, D. P. Huttenlocher, K. Kedem, and J. S. B. Mitchell. An efficiently computable metric for comparing polygonal shapes. IEEE Trans. PAMI, 13(3):209-216, 1991. ACI92 D. Alberts, G. Cattaneo, and G. Italiano. An empirical study of dynamic graph algorithms. In Proc. Seventh ACM-SIAM Symp. Discrete Algorithms (SODA), pages 192-201, 1992. ADKF70 V. Arlazarov, E. Dinic, M. Kronrod, and I. Faradzev. On economical construction of the transitive closure of a directed graph. Soviet Mathematics, Doklady, 11:1209-1210, 1970. Adl94a

file:///E|/BOOK/BOOK5/NODE232.HTM (1 of 50) [19/1/2003 1:28:01]

References

L. Adleman. Algorithmic number theory - the complexity contribution. In Proc. 35th IEEE Symp. Foundations of Computer Science (FOCS), pages 88-113, 1994. Adl94b L. M. Adleman. Molecular computations of solutions to combinatorial problems. Science, 266:1021-1024, November 11, 1994. AE83 D. Avis and H. ElGindy. A combinatorial approach to polygon similarity. IEEE Trans. Inform. Theory, IT-2:148-150, 1983. AF92 D. Avis and K. Fukuda. A pivoting algorithm for convex hulls and vertex enumeration of arrangements and polyhedra. Discrete Comput. Geom., 8:295-313, 1992. AG86 A. Apostolico and R. Giancarlo. The Boyer-Moore-Galil string searching strategies revisited. SIAM J. Computing, 15:98-105, 1986. AGSS89 A. Aggarwal, L. Guibas, J. Saxe, and P. Shor. A linear-time algorithm for computing the Voronoi diagram of a convex polygon. Discrete and Computational Geometry, 4:591-604, 1989. AGU72 A. Aho, M. Garey, and J. Ullman. The transitive reduction of a directed graph. SIAM J. Computing, 1:131-137, 1972. Aho90 A. Aho. Algorithms for finding patterns in strings. In J. van Leeuwen, editor, Handbook of Theoretical Computer Science: Algorithms and Complexity, volume A, pages 255-300. MIT Press, 1990. AHU74 A. Aho, J. Hopcroft, and J. Ullman. The Design and Analysis of Computer Algorithms. AddisonWesley, Reading MA, 1974. AHU83 A. Aho, J. Hopcroft, and J. Ullman. Data Structures and Algorithms. Addison-Wesley, Reading MA, 1983. Aig88 M. Aigner. Combinatorial Search. Wiley-Teubner, 1988. file:///E|/BOOK/BOOK5/NODE232.HTM (2 of 50) [19/1/2003 1:28:01]

References

AK89 E. Aarts and J. Korst. Simulated annealing and Boltzman machines: A stochastic approach to combinatorial optimization and neural computing. John Wiley and Sons, 1989. AKD83 J. H. Ahrens, K. D. Kohrt, and U. Dieter. Sampling from gamma and Poisson distributions. ACM Trans. Math. Softw., 9(2):255-257, June 1983. AM93 S. Arya and D. Mount. Approximate nearest neighbor queries in fixed dimensions. In Proc. Fourth ACM-SIAM Symp. Discrete Algorithms (SODA), pages 271-280, 1993. AMN 94 S. Arya, D. Mount, N. Netanyahu, R. Silverman, and A. Wu. Approximate nearest neighbor queries in fixed dimensions. In Proc. Fifth ACM-SIAM Symp. Discrete Algorithms (SODA), pages 573-582, 1994. AMO93 R. Ahuja, T. Magnanti, and J. Orlin. Network Flows. Prentice Hall, Englewood Cliffs NJ, 1993. AMOT88 R. Ahuja, K. Mehlhorn, J. Orlin, and R. Tarjan. Faster algorithms for the shortest path problem. Technical Report 193, MIT Operations Research Center, 1988. AMWW88 H. Alt, K. Mehlhorn, H. Wagener, and E. Welzl. Congruence, similarity and symmetries of geometric objects. Discrete Comput. Geom., 3:237-256, 1988. And76 G. Andrews. The Theory of Partitions. Addison-Wesley, Reading, Mass., 1976. Apo85 A. Apostolico. The myriad virtues of subword trees. In A. Apostolico and Z. Galil, editors, Combinatorial algorithms on words. Springer-Verlag, 1985. APT79 B. Aspvall, M. Plass, and R. Tarjan. A linear-time algorithm for testing the truth of certain quantified boolean formulas. Info. Proc. Letters, 8:121-123, 1979. Aro96 S. Arora. Polynomial time approximations schemes for Euclidean TSP and other geometric

file:///E|/BOOK/BOOK5/NODE232.HTM (3 of 50) [19/1/2003 1:28:01]

References

problems. In Proc. 37th IEEE Foundations of Computer Science (FOCS '96), pages 1-10, 1996. AS89 C. Aragon and R. Seidel. Randomized search trees. In Proc. 30th IEEE Symp. Foundations of Computer Science, pages 540-545, 1989. Ata83 M. Atallah. A linear time algorithm for the Hausdorff distance between convex polygons. Info. Proc. Letters, 8:207-209, 1983. Ata84 M. Atallah. Checking similarity of planar figures. Internat. J. Comput. Inform. Sci., 13:279-290, 1984. Aur91 F. Aurenhammer. Voronoi diagrams: a survey of a fundamental data structure. ACM Computing Surveys, 23:345-405, 1991. Baa88 S. Baase. Computer Algorithms. Addison-Wesley, Reading MA, second edition, 1988. BCGR92 D. Berque, R. Cecchini, M. Goldberg, and R. Rivenburgh. The SetPlayer system for symbolic computation on power sets. J. Symbolic Computation, 14:645-662, 1992. BCW90 T. Bell, J. Cleary, and I. Witten. Text Compression. Prentice Hall, Englewood Cliffs NJ, 1990. BDH97 C. Barber, D. Dobkin, and H. Huhdanpaa. The Quickhull algorithm for convex hulls. ACM Trans. on Mathematical Software, 22:469-483, 1997. Bel58 R. Bellman. On a routing problem. Quarterly of Applied Mathematics, 16:87-90, 1958. Ben75 J. L. Bentley. Multidimensional binary search trees used for associative searching. Communications of the ACM, 18:509-517, 1975. Ben86 J. Bentley. Programming Pearls. Addison-Wesley, Reading MA, 1986.

file:///E|/BOOK/BOOK5/NODE232.HTM (4 of 50) [19/1/2003 1:28:01]

References

Ben90 J. Bentley. More Programming Pearls. Addison-Wesley, Reading MA, 1990. Ben92a J. Bentley. Fast algorithms for geometric traveling salesman problems. ORSA J. Computing, 4:387411, 1992. Ben92b J. L. Bentley. Software exploratorium: The trouble with qsort. UNIX Review, 10(2):85-93, February 1992. Ber89 C. Berge. Hypergraphs. North-Holland, Amsterdam, 1989. BETT94 G. Di Battista, P. Eades, R. Tamassia, and I. Tollis. Algorithms for drawing graphs: An annotated bibliography. Computational Geometry: Theory and Applications, 4, 1994. BFP 72 M. Blum, R. Floyd, V. Pratt, R. Rivest, and R. Tarjan. Time bounds for selection. J. Computer and System Sciences, 7:448-461, 1972. BG95 J. Berry and M. Goldberg. Path optimization and near-greedy analysis for graph partitioning: An empirical study. In Proc. 6th ACM-SIAM Symposium on Discrete Algorithms, pages 223-232, 1995. BGL 95 C. Di Battista, A. Garg, G. Liotta, R. Tamassia, E. Tassinari, and F. Vargiu. An experimental comparison of three graph drawing algorithms. In Proc. 11th ACM Symposium on Computational Geometry, pages 306-315, 1995. BGS95 M Bellare, O. Goldreich, and M. Sudan. Free bits, PCPs, and non-approximability - towards tight results. In Proc. IEEE 36th Symp. Foundations of Computer Science, pages 422-431, 1995. BH90 F. Buckley and F. Harary. Distances in Graphs. Addison-Wesley, Redwood City, Calif., 1990. BJ85 S. Bent and J. John. Finding the median requires 2n comparisons. In Proc. 17th ACM Symp. on Theory of Computing, pages 213-216, 1985. file:///E|/BOOK/BOOK5/NODE232.HTM (5 of 50) [19/1/2003 1:28:01]

References

BJL 91 A. Blum, T. Jiang, M. Li, J. Tromp, and M. Yanakakis. Linear approximation of shortest superstrings. In Proc. 23rd ACM Symp. on Theory of Computing., pages 328-336, 1991. BJLM83 J. Bentley, D. Johnson, F. Leighton, and C. McGeoch. An experimental study of bin packing. In Proc. 21st Allerton Conf. on Communication, Control, and Computing, pages 51-60, 1983. BL77 B. P. Buckles and M. Lybanon. Generation of a vector from the lexicographical index. ACM Trans. Math. Softw., 3(2):180-182, June 1977. BLS91 D. Bailey, K. Lee, and H. Simon. Using Strassen's algorithm to accelerate the solution of linear systems. J. Supercomputing, 4:357-371, 1991. Blu67 H. Blum. A transformation for extracting new descriptions of shape. In W. Wathen-Dunn, editor, Models for the Perception of speech and Visual Form, pages 362-380. MIT Press, 1967. BLW76 N. L. Biggs, E. K. Lloyd, and R. J. Wilson. Graph Theory 1736-1936. Clarendon Press, Oxford, 1976. BM53 G. Birkhoff and S. MacLane. A survey of modern algebra. Macmillian, New York, 1953. BM72 R. Bayer and E. McCreight. Organization and maintenance of large ordered indexes. Acta Informatica, 1:173-189, 1972. BM77 R. Boyer and J. Moore. A fast string-searching algorithm. Communications of the ACM, 20:762772, 1977. BM89 J. Boreddy and R. N. Mukherjee. An algorithm to find polygon similarity. Inform. Process. Lett., 33(4):205-206, 1989. BO79 J. Bentley and T. Ottmann. Algorithms for reporting and counting geometric intersections. IEEE

file:///E|/BOOK/BOOK5/NODE232.HTM (6 of 50) [19/1/2003 1:28:01]

References

Transactions on Computers, C-28:643-647, 1979. BO83 M. Ben-Or. Lower bounds for algebraic computation trees. In Proc. Fifteenth ACM Symp. on Theory of Computing, pages 80-86, 1983. BP76 E. Balas and M. Padberg. Set partitioning - a survey. SIAM Review, 18:710-760, 1976. BR80 I. Barrodale and F. D. K. Roberts. Solution of the constrained ACM Trans. Math. Softw., 6(2):231-235, June 1980.

linear approximation problem.

BR95 A. Binstock and J. Rex. Practical Algorithms for Programmers. Addison-Wesley, Reading MA, 1995. Bre73 R. Brent. Algorithms for minimization without derivatives. Prentice-Hall, Englewood Cliffs NJ, 1973. Bre74 R. P. Brent. A Gaussian pseudo-random number generator. Commun. ACM, 17(12):704-706, December 1974. Brè79 D. Brèlaz. New methods to color the vertices of a graph. Communications of the ACM, 22:251256, 1979. Bri74 E. Brigham. The Fast Fourier Transform. Prentice Hall, Englewood Cliffs NJ, 1974. Bro74 F. Brooks. The Mythical Man-Month. Addison-Wesley, Reading MA, 1974. Bro88 R. Brown. Calendar queuing method for future event list manipulation. Comm. ACM, 30, 1988. Brz64 J. Brzozowski. Derivatives of regular expressions. J. ACM, 11:481-494, 1964. BS76 file:///E|/BOOK/BOOK5/NODE232.HTM (7 of 50) [19/1/2003 1:28:01]

References

J. Bentley and M. Shamos. Divide-and-conquer in higher-dimensional space. In Proc. Eighth ACM Symp. Theory of Computing, pages 220-230, 1976. BS81 I. Barrodale and G. F. Stuart. A Fortran program for solving 7(3):391-397, September 1981.

. ACM Trans. Math. Softw.,

BS86 G. Berry and R. Sethi. From regular expressions to deterministic automata. Theoretical Computer Science, 48:117-126, 1986. BS93 E. Biham and A. Shamir. Differential Cryptanalysis of the Data Encryption Standard. SpringerVerlag, Berlin, 1993. BS96 E. Bach and J. Shallit. Algorithmic Number Theory: Efficient Algorithms, volume 1. MIT Press, Cambridge MA, 1996. BS97 R. Bradley and S. Skiena. Fabricating arrays of strings. In Proc. First Int. Conf. Computational Molecular Biology (RECOMB '97), pages 57-66, 1997. BT92 J. Buchanan and P. Turner. Numerical methods and analysis. McGraw-Hill, New York, 1992. Buc94 A. G. Buckley. A Fortran 90 code for unconstrained nonlinear minimization. ACM Trans. Math. Softw., 20(3):354-372, September 1994. BW91 G. Brightwell and P. Winkler. Counting linear extensions is #P-complete. In Proc. 23rd ACM Symp. Theory Computing (STOC), pages 175-181, 1991. BYGNR94 R. Bar-Yehuda, D. Geiger, J. Naor, and R. Roth. Approximation algorithms for the vertex feedback set problem with applications to constraint satisfaction and Bayesian inference. In Proc. Fifth ACM-SIAM Symp. Discrete Algorithms, pages 344-354, 1994. Can87 J. Canny. The complexity of robot motion planning. MIT Press, Cambridge MA, 1987.

file:///E|/BOOK/BOOK5/NODE232.HTM (8 of 50) [19/1/2003 1:28:01]

References

CC92 S. Carlsson and J. Chen. The complexity of heaps. In Proc. Third ACM-SIAM Symp. on Discrete Algorithms, pages 393-402, 1992. CCDG82 P. Chinn, J. Chvátolvá, A. K. Dewdney, and N. E. Gibbs. The bandwidth problem for graphs and matrices - a survey. J. Graph Theory, 6:223-254, 1982. CD85 B. Chazelle and D. Dobkin. Optimal convex decompositions. In G. Toussaint, editor, Computational Geometry, pages 63-133. North-Holland, Amsterdam, 1985. CDT95 G. Carpento, M. Dell'Amico, and P. Toth. CDT: A subroutine for the exact solution of large-scale, asymmetric traveling salesman problems. ACM Trans. Math. Softw., 21(4):410-415, December 1995. CE92 B. Chazelle and H. Edelsbrunner. An optimal algorithm for intersecting line segments. J. ACM, 39:1-54, 1992. CG94 B. Cherkassky and A. Goldberg. On implementing push-relabel method for the maximum flow problem. Technical Report 94-1523, Department of Computer Science, Stanford University, 1994. CGJ96 E. G. Coffman, M. R. Garey, and D. S. Johnson. Approximation algorithms for bin packing: a survey. In D. Hochbaum, editor, Approximation algorithms. PWS Publishing, 1996. CGL85 B. Chazelle, L. Guibas, and D. T. Lee. The power of geometric duality. BIT, 25:76-90, 1985. CGPS76 H. L. Crane Jr., N. F. Gibbs, W. G. Poole Jr., and P. K. Stockmeyer. Matrix bandwidth and profile reduction. ACM Trans. Math. Softw., 2(4):375-377, December 1976. CGR93 B. Cherkassky, A. Goldberg, and T. Radzik. Shortest paths algorithms: theory and experimental evaluation. Technical Report 93-1480, Department of Computer Science, Stanford University, 1993. Cha71 file:///E|/BOOK/BOOK5/NODE232.HTM (9 of 50) [19/1/2003 1:28:01]

References

J. M. Chambers. Partial sorting. Commun. ACM, 14(5):357-358, May 1971. Cha91 B. Chazelle. Triangulating a simple polygon in linear time. Discrete and Computational Geometry, 6:485-524, 1991. Che80 T-Y. Cheung. A program for the multifacility location problem with rectilinear distance by the minimum-cut approach. ACM Trans. Math. Softw., 6(3):430-431, September 1980. Chr76 N. Christofides. Worst-case analysis of a new heuristic for the traveling salesman problem. Technical report, Graduate School of Industrial Administration, Carnegie-Mellon University, Pittsburgh PA, 1976. CHT90 J. Cai, X. Han, and R. Tarjan. New solutions to four planar graph problems. Technical report, New York University, 1990. Chv83 V. Chvatal. Linear Programming. Freeman, San Francisco, 1983. CK70 D. Chand and S. Kapur. An algorithm for convex polytopes. J. ACM, 17:78-86, 1970. CK75 N. Christofides and S. Korman. A computational survey of methods for the set covering problem. Management Science, 21:591-599, 1975. CK80 W. Cheney and D. Kincaid. Numerical Mathematics and Computing. Brooks/Cole, Monterey CA, 1980. CK94 A. Chetverin and F. Kramer. Oligonucleotide arrays: New concepts and possibilities. Bio/Technology, 12:1093-1099, 1994. Cla92 K. L. Clarkson. Safe and effective determinant evaluation. In Proc. 31st IEEE Symposium on Foundations of Computer Science, pages 387-395, Pittsburgh, PA, 1992. CLR90 file:///E|/BOOK/BOOK5/NODE232.HTM (10 of 50) [19/1/2003 1:28:01]

References

T. Cormen, C. Leiserson, and R. Rivest. Introduction to Algorithms. MIT Press, Cambridge MA, 1990. CM69 E. Cuthill and J. McKee. Reducing the bandwidth of sparse symmetric matrices. In Proc. 24th Nat. Conf. ACM, pages 157-172, 1969. Cof76 E. Coffman. Computer and Job shop Scheduling. Wiley, New York, 1976. Coh94 E. Cohen. Estimating the size of the transitive closure in linear time. In 35th Annual Symposium on Foundations of Computer Science, pages 190-200. IEEE, 1994. Coo71 S. Cook. The complexity of theorem proving procedures. In Proc. Third ACM Symp. Theory of Computing, pages 151-158, 1971. CP90 R. Carraghan and P. Paradalos. An exact algorithm for the maximum clique problem. In Operations Research Letters, volume 9, pages 375-382, 1990. CR76 J. Cohen and M. Roth. On the implementation of Strassen's fast multiplication algorithm. Acta Informatica, 6:341-355, 1976. CR94 M. Crochemore and W. Rytter. Text Algorithms. Oxford University Press, New York, 1994. Cra94 R. Crandall. Topics in Scientific Computation. Telos/Springer-Verlag, New York, 1994. CS93 J. Conway and N. Sloane. Sphere packings, lattices, and groups. Springer-Verlag, New York, 1993. CT65 J. Cooley and J. Tukey. An algorithm for the machine calculation of complex Fourier series. Mathematics of Computation, 19:297-301, 1965. CT71 M. W. Coleman and M. S. Taylor. Circular integer partitioning. Commun. ACM, 14(1):48, January file:///E|/BOOK/BOOK5/NODE232.HTM (11 of 50) [19/1/2003 1:28:01]

References

1971. CT80 G. Carpaneto and P. Toth. Solution of the assignment problem. ACM Trans. Math. Softw., 6(1):104111, March 1980. CT92 Y. Chiang and R. Tamassia. Dynamic algorithms in computational geometry. Proc. IEEE, 80:14121434, 1992. CW79 B. Commentz-Walter. A string matching algorithm fast on the average. In Proc. Sixth Int. Coll. on Automata, Languages, and Programming (ICALP), pages 118-132. Springer Verlag, Lecture Notes in Computer Science, 1979. CW87 D. Coppersmith and S. Winograd. Matrix multiplication via arithmetic progressions. In Proc. Nineteenth ACM Symp. Theory of Computing, pages 1-6, 1987. Dan63 G. Dantzig. Linear programming and extensions. Princeton University Press, Princeton NJ, 1963. Dau92 I. Daubechies. Ten Lectures on Wavelets. SIAM, Philadelphia, 1992. DB74 G. Dahlquist and A. Bjorck. Numerical Methods. Prentice-Hall, Englewood Cliffs NJ, 1974. DB86 G. Davies and S. Bowsher. Algorithms for pattern matching. Software - Practice and Experience, 16:575-601, 1986. Den82 D. Denning. Cryptography and Data Security. Addison-Wesley, Reading MA, 1982. DF79 E. Denardo and B. Fox. Shortest-route methods: 1. reaching, pruning, and buckets. Operations Research, 27:161-186, 1979. DFJ54 G. Dantzig, D. Fulkerson, and S. Johnson. Solution of a large-scale traveling-salesman problem. Operations Research, 2:393-410, 1954. file:///E|/BOOK/BOOK5/NODE232.HTM (12 of 50) [19/1/2003 1:28:01]

References

dFPP88 H. de Fraysseix, J. Pach, and R. Pollack. Small sets supporting Fary embeddings of planar graphs. In Proc. of the 20th Symposium on the Theory of Computing, pages 426-433. ACM, 1988. DG87 J. Dongarra and E. Grosse. Distribution of mathematical software via electronic mail. Communications of the ACM, 30:403-407, 1987. DGKK79 R. Dial, F. Glover, D. Karney, and D. Klingman. A computational analysis of alternative algorithms and labeling techniques for finding shortest path trees. Networks, 9:215-248, 1979. DH73 R. Duda and P. Hart. Pattern Classification and Scene Analysis. Wiley-Interscience, New York, 1973. DH92 D. Du and F. Hwang. A proof of Gilbert and Pollak's conjecture on the Steiner ratio. Algorithmica, 7:121-135, 1992. Dij59 E. W. Dijkstra. A note on two problems in connection with graphs. Nuerische Mathematik, 1:269271, 1959. DJ92 G. Das and D. Joseph. Minimum vertex hulls for polyhedral domains. Theoret. Comput. Sci., 103:107-135, 1992. DL76 D. Dobkin and R. Lipton. Multidimensional searching problems. SIAM J. Computing, 5:181-186, 1976. DLR79 D. Dobkin, R. Lipton, and S. Reiss. Linear programming is log-space hard for P. Info. Processing Letters, 8:96-97, 1979. DM80 D. Dobkin and J. I. Munro. Determining the mode. Theoretical Computer Science, 12:255-263, 1980. DM97

file:///E|/BOOK/BOOK5/NODE232.HTM (13 of 50) [19/1/2003 1:28:01]

References

K. Daniels and V. Milenkovic. Multiple translational containment. part I: an approximation algorithm. Algorithmica, 1997. DMBS79 J. Dongarra, C. Moler, J. Bunch, and G. Stewart. LINPACK User's Guide. SIAM Publications, Philadelphia, 1979. DNRT81 J. J. Ducroz, S. M. Nugent, J. K. Reid, and D. B. Taylor. Solution of real linear equations in a paged virtual store. ACM Trans. Math. Softw., 7(4):537-541, December 1981. Dor94 S. Dorward. A survey of object-space hidden surface removal. Int. J. Computational Geometry Theory and Applications, 4:325-362, 1994. DP73 D. H. Douglas and T. K. Peucker. Algorithms for the reduction of the number of points required to represent a digitized line or its caricature. Canadian Cartographer, 10(2):112-122, December 1973. DP84 N. Deo and C. Pang. Shortest path algorithms: Taxonomy and annotation. Networks, 14:275-323, 1984. DR90 N. Dershowitz and E. Reingold. Calendrical calculations. Software - Practice and Experience, 20:899-928, 1990. DR97 N. Dershowitz and E. Reingold. Calendrical Calculations. Cambridge University Press, New York, 1997. DRR 95 S. Dawson, C.R. Ramakrishnan, I.V. Ramakrishnan, K. Sagonas, S. Skiena, T. Swift, and D.S. Warren. Unification factoring for efficient execution of logic programs. In 22nd ACM Symposium on Principles of Programming Languages (POPL '95), pages 247-258, 1995. DS88 D. Dobkin and D. Silver. Recipes for geometry and numerical analysis. In Proc. 4th ACM Symp. Computational Geometry, pages 93-105, 1988. Duf81 file:///E|/BOOK/BOOK5/NODE232.HTM (14 of 50) [19/1/2003 1:28:01]

References

S. Duff. Permutations for a zero-free diagonal. ACM Trans. Math. Softw., 7(3):387-390, September 1981. dVS82 G. de V. Smit. A comparison of three string matching algorithms. Software - Practice and Experience, 12:57-66, 1982. DZ95 D. Dor and U. Zwick. Selecting the median. In Proc. Sixth ACM-SIAM Symp. Discrete Algorithms (SODA), pages 28-37, 1995. Ebe88 J. Ebert. Computing Eulerian trails. Info. Proc. Letters, 28:93-97, 1988. Edd77 W. F. Eddy. CONVEX: a new convex hull algorithm for planar sets. ACM Trans. Math. Softw., 3(4):411-412, December 1977. Ede87 H. Edelsbrunner. Algorithms for Combinatorial Geometry. Springer-Verlag, Berlin, 1987. Edm65 J. Edmonds. Paths, trees, and flowers. Canadian J. Math., 17:449-467, 1965. Edm71 J. Edmonds. Matroids and the greedy algorithm. Mathematical Programming, 1:126-136, 1971. EG60 P. Erdös and T. Gallai. Graphs with prescribed degrees of vertices. Mat. Lapok (Hungarian), 11:264-274, 1960. EG89 H. Edelsbrunner and L. Guibas. Topologically sweeping an arrangement. J. Computer and System Sciences, 38:165-194, 1989. EG91 H. Edelsbrunner and L. Guibas. Corrigendum: Topologically sweeping an arrangement. J. Computer and System Sciences, 42:249-251, 1991. EGIN92 D. Eppstein, Z. Galil, G. F. Italiano, and A. Nissenzweig. Sparsification: A technique for speeding up dynamic graph algorithms. In Proc. 33rd IEEE Symp. on Foundations of Computer Science file:///E|/BOOK/BOOK5/NODE232.HTM (15 of 50) [19/1/2003 1:28:01]

References

(FOCS), pages 60-69, 1992. EGS86 H. Edelsbrunner, L. Guibas, and J. Stolfi. Optimal point location in a monotone subdivision. SIAM J. Computing, 15:317-340, 1986. EJ73 J. Edmonds and E. Johnson. Matching, Euler tours, and the Chinese postman. Math. Programming, 5:88-124, 1973. EKA84 M. I. Edahiro, I. Kokubo, and T. Asano. A new point location algorithm and its practical efficiency - comparison with existing algorithms. ACM Trans. Graphics, 3:86-109, 1984. EKS83 H. Edelsbrunner, D. Kirkpatrick, and R. Seidel. On the shape of a set of points in the plane. IEEE Trans. on Information Theory, IT-29:551-559, 1983. ELS93 P. Eades, X. Lin, and W. F. Smyth. A fast and effective heuristic for the feedback arc set problem. Info. Proc. Letters, 47:319-323, 1993. EM94 H. Edelsbrunner and Ernst P. Mücke. Three-dimensional alpha shapes. ACM Transactions on Graphics, 13:43-72, 1994. ES74 P. Erdös and J. Spencer. Probabilistic Methods in Combinatorics. Academic Press, New York, 1974. ES86 H. Edelsbrunner and R. Seidel. Voronoi diagrams and arrangements. Discrete and Computational Geometry, 1:25-44, 1986. ESS93 H. Edelsbrunner, R. Seidel, and M. Sharir. On the zone theorem for hyperplane arrangements. SIAM J. Computing, 22:418-429, 1993. ESV96 F. Evans, S. Skiena, and A. Varshney. Optimizing triangle strips for fast rendering. In Proc. IEEE Visualization '96, pages 319-326, 1996.

file:///E|/BOOK/BOOK5/NODE232.HTM (16 of 50) [19/1/2003 1:28:02]

References

Eul36 L. Euler. Solutio problematis ad geometriam situs pertinentis. Commentarii Academiae Scientiarum Petropolitanae, 8:128-140, 1736. Eve79a S. Even. Graph Algorithms. Computer Science Press, Rockville MD, 1979. Eve79b G. Everstine. A comparison of three resequencing algorithms for the reduction of matrix profile and wave-front. Int. J. Numerical Methods in Engr., 14:837-863, 1979. F48 I. Fáry. On straight line representation of planar graphs. Acta. Sci. Math. Szeged, 11:229-233, 1948. Faj87 S. Fajtlowicz. On conjectures of Graffiti. Discrete Mathematics, 72:113-118, 1987. FF62 L. Ford and D. R. Fulkerson. Flows in Networks. Princeton University Press, Princeton NJ, 1962. Fis95 P. Fishwick. Simulation Model Design and Execution: Building Digital Worlds. Prentice Hall, Englewood Cliffs, NJ, 1995. FJ95 A. Frieze and M. Jerrum. An analysis of a Monte Carlo algorithm for estimating the permanent. Combinatorica, 15, 1995. Fle74 H. Fleischner. The square of every two-connected graph is Hamiltonian. J. Combinatorial Theory, B, 16:29-34, 1974. Fle80 R. Fletcher. Practical Methods of Optimization: Unconstrained Optimization, volume 1. John Wiley, Chichester, 1980. Flo62 R. Floyd. Algorithm 97 (shortest path). Communications of the ACM, 7:345, 1962. Flo64 R. Floyd. Algorithm 245 (treesort). Communications of the ACM, 18:701, 1964. file:///E|/BOOK/BOOK5/NODE232.HTM (17 of 50) [19/1/2003 1:28:02]

References

FM71 M. Fischer and A. Meyer. Boolean matrix multiplication and transitive closure. In IEEE 12th Symp. on Switching and Automata Theory, pages 129-131, 1971. For87 S. Fortune. A sweepline algorithm for Voronoi diagrams. Algorithmica, 2:153-174, 1987. For92 S. Fortune. Voronoi diagrams and Delaunay triangulations. In D.-Z. Du and F. Hwang, editors, Computing in Euclidean Geometry, volume 1, pages 193-234. World Scientific, 1992. FP75a D. Fayard and G. Plateau. Resolution of the 0-1 knapsack problem: Comparison of methods. Math. Programming, 8:272-307, 1975. FP75b H. Feng and T. Pavlidis. Decomposition of polygons into simpler components: feature generation for syntactic pattern recognition. IEEE Transactions on Computers, C-24:636-650, 1975. FR75 R. Floyd and R. Rivest. Expected time bounds for selection. Communications of the ACM, 18:165172, 1975. FR94 M. Fürer and B. Raghavachari. Approximating the minimum-degree Steiner tree to within one of optimal. J. Algorithms, 17:409-423, 1994. Fra79 D. Fraser. An optimized mass storage FFT. ACM Trans. Math. Softw., 5(4):500-517, December 1979. Fre62 E. Fredkin. Trie memory. Communications of the ACM, 3:490-499, 1962. FT87 M. Fredman and R. Tarjan. Fibonacci heaps and their uses in improved network optimization algorithms. J. ACM, 34:596-615, 1987. Fuj96 T. Fujito. A note on approximation of the vertex cover and feedback vertex set problems. Info. Proc. Letters, 59:59-63, 1996.

file:///E|/BOOK/BOOK5/NODE232.HTM (18 of 50) [19/1/2003 1:28:02]

References

FvW93 S. Fortune and C. van Wyk. Efficient exact arithmetic for computational geometry. In Proc. 9th ACM Symp. Computational Geometry, pages 163-172, 1993. FW77 S. Fiorini and R. Wilson. Edge-colourings of graphs. Research Notes in Mathematics 16, Pitman, London, 1977. FW93 M. Fredman and D. Willard. Surpassing the information theoretic bound with fusion trees. J. Computer and System Sci., 47:424-436, 1993. Gab76 H. Gabow. An efficient implementation of Edmond's algorithm for maximum matching on graphs. J. ACM, 23:221-234, 1976. Gab77 H. Gabow. Two algorithms for generating weighted spanning trees in order. SIAM J. Computing, 6:139-150, 1977. Gal86 Z. Galil. Efficient algorithms for finding maximum matchings in graphs. ACM Computing Surveys, 18:23-38, 1986. GBDS80 B. Golden, L. Bodin, T. Doyle, and W. Stewart. Approximate traveling salesman algorithms. Operations Research, 28:694-711, 1980. GBY91 G. Gonnet and R. Baeza-Yates. Handbook of Algorithms and Data Structures. Addison-Wesley, Wokingham, England, second edition, 1991. GGJ77 M. Garey, R. Graham, and D. Johnson. The complexity of computing Steiner minimal trees. SIAM J. Appl. Math., 32:835-859, 1977. GGJK78 M. Garey, R. Graham, D. Johnson, and D. Knuth. Complexity results for bandwidth minimization. SIAM J. Appl. Math., 34:477-495, 1978. GH85 R. Graham and P. Hell. On the history of the minimum spanning tree problem. Annals of the file:///E|/BOOK/BOOK5/NODE232.HTM (19 of 50) [19/1/2003 1:28:02]

References

History of Computing, 7:43-57, 1985. GHMS93 L. J. Guibas, J. E. Hershberger, J. S. B. Mitchell, and J. S. Snoeyink. Approximating polygons and subdivisions with minimum link paths. Internat. J. Comput. Geom. Appl., 3(4):383-415, December 1993. GHR95 R. Greenlaw, J. Hoover, and W. Ruzzo. Limits to Parallel Computation: P-completeness theory. Oxford University Press, New York, 1995. GI89 D. Gusfield and R. Irving. The Stable Marriage Problem: structure and algorithms. MIT Press, Cambridge MA, 1989. GI91 Z. Galil and G. Italiano. Data structures and algorithms for disjoint set union problems. ACM Computing Surveys, 23:319-344, 1991. Gib76 N. E. Gibbs. A hybrid profile reduction algorithm. ACM Trans. Math. Softw., 2(4):378-387, December 1976. GJ77 M. Garey and D. Johnson. The rectilinear Steiner tree problem is NP-complete. SIAM J. Appl. Math., 32:826-834, 1977. GJ79 M. R. Garey and D. S. Johnson. Computers and Intractability: A Guide to the theory of NPcompleteness. W. H. Freeman, San Francisco, 1979. GJPT78 M. Garey, D. Johnson, F. Preparata, and R. Tarjan. Triangulating a simple polygon. Info. Proc. Letters, 7:175-180, 1978. GK79 A. Goralcikiova and V. Konbek. A reduct and closure algorithm for graphs. In Mathematical Foundations of Computer Science, pages 301-307. Springer Verlag, Lecture Notes in Computer Science V. 74, 1979. GK93 A. Goldberg and R. Kennedy. An efficient cost scaling algorithm for the assignment problem. file:///E|/BOOK/BOOK5/NODE232.HTM (20 of 50) [19/1/2003 1:28:02]

References

Technical Report 93-1481, Department of Computer Science, Stanford University, 1993. GKK74 F. Glover, D. Karney, and D. Klingman. Implementation and computational comparisons of primaldual computer codes for minimum-cost network flow problems. Networks, 4:191-212, 1974. GKP89 R. Graham, D. Knuth, and O. Patashnik. Concrete Mathematics. Addison-Wesley, Reading MA, 1989. Glo89a F. Glover. Tabu search - Part 1. ORSA Journal on Computing, 1(3):190-206, 1989. Glo89b F. Glover. Tabu search - Part 2. ORSA Journal on Computing, 2(1):4-32, 1989. Glo90 F. Glover. Tabu search: A tutorial. Interfaces, Vol. 20, :4, 74-94, 1990. GM86 G. Gonnet and J.I. Munro. Heaps on heaps. SIAM J. Computing, 15:964-971, 1986. Gol89 D. E. Goldberg. Genetic Algorithms in Search, Optimization, and Machine Learning. AddisonWesley, 1989. Gol92 A. Goldberg. An efficient implementation of a scaling minimum-cost flow algorithm. Technical Report 92-1439, Department of Computer Science, Stanford University, 1992. Gol93 L. Goldberg. Efficient Algorithms for Listing Combinatorial Structures. Cambridge University Press, 1993. GP68 E. Gilbert and H. Pollak. Steiner minimal trees. SIAM J. Applied Math., 16:1-29, 1968. GP79 B. Gates and C. Papadimitriou. Bounds for sorting by prefix reversals. Discrete Mathematics, 27:47-57, 1979. GPS76 file:///E|/BOOK/BOOK5/NODE232.HTM (21 of 50) [19/1/2003 1:28:02]

References

N. Gibbs, W. Poole, and P. Stockmeyer. A comparison of several bandwidth and profile reduction algorithms. ACM Trans. Math. Software, 2:322-330, 1976. Gra53 F. Gray. Pulse code communication. US Patent 2632058, March 17, 1953. Gra72 R. Graham. An efficient algorithm for determining the convex hull of a finite planar point set. Info. Proc. Letters, 1:132-133, 1972. GS62 D. Gale and L. Shapely. College admissions and the stability of marriages. American Math. Monthly, 69:9-14, 1962. GS78 L. Guibas and R. Sedgewick. A dichromatic framework for balanced trees. In Proc. 19th IEEE Symp. Foundations of Computer Science, pages 8-21, 1978. GT88 A. Goldberg and R. Tarjan. A new approach to the maximum flow problem. J. ACM, pages 921940, 1988. GT89 H. Gabow and R. Tarjan. Faster scaling algorithms for network problems. SIAM J. Computing, 18:1013-1036, 1989. Gup66 R. P. Gupta. The chromatic index and the degree of a graph. Notices of the Amer. Math. Soc., 13:66T-429, 1966. Gus97 D. Gusfield. String Algorithms. Cambridge University Press, 1997. GW95 M. Goemans and D. Williamson. .878-approximation algorithms for MAX CUT and MAX 2SAT. J. ACM, 42:1115-1145, 1995. HD80 P. Hall and G. Dowling. Approximate string matching. ACM Computing Surveys, 12:381-402, 1980. HG95 file:///E|/BOOK/BOOK5/NODE232.HTM (22 of 50) [19/1/2003 1:28:02]

References

P. Heckbert and M. Garland. Fast polygonal approximation of terrains and height fields. Technical Report CMU-CS-95-181, School of Computer Science, Carnegie-Mellon University, 1995. Him94 M. Himsolt. GraphEd: A graphical platform for the implementation of graph algorithms. In Proc. Graph Drawing (GD '94), volume 894, pages 182-193, 1994. Hir75 D. Hirschberg. A linear-space algorithm for computing maximum common subsequences. Communications of the ACM, 18:341-343, 1975. HJS84 R. E. Haymond, J. P. Jarvis, and D. R. Shier. Minimum spanning tree for moderate integer weights. ACM Trans. Math. Softw., 10(1):108-111, March 1984. HK73 J. Hopcroft and R. Karp. An Computing, 2:225-231, 1973.

algorithm for maximum matchings in bipartite graphs. SIAM J.

HK90 D. P. Huttenlocher and K. Kedem. Computing the minimum Hausdorff distance for point sets under translation. In Proc. 6th Annu. ACM Sympos. Comput. Geom., pages 340-349, 1990. HM83 S. Hertel and K. Mehlhorn. Fast triangulation of simple polygons. In Proc. 4th Internat. Conf. Found. Comput. Theory, pages 207-218. Lecture Notes in Computer Science, Vol. 158, 1983. HMMN84 S. Hertel, K. Mehlhorn, M. Mäntylä, and J. Nievergelt. Space sweep solves intersection of two convex polyhedra elegantly. Acta Informatica, 21:501-519, 1984. Hoa61 C. A. R. Hoare. Algorithm 63 (partition) and algorithm 65 (find). Communications of the ACM, 4:321-322, 1961. Hoa62 C. A. R. Hoare. Quicksort. Computer Journal, 5:10-15, 1962. Hoc96 D. Hochbaum, editor. Approximation Algorithms for NP-hard Problems. PWS Publishing, Boston, 1996.

file:///E|/BOOK/BOOK5/NODE232.HTM (23 of 50) [19/1/2003 1:28:02]

References

Hof82 C. M. Hoffmann. Group-theoretic algorithms and graph isomorphism, volume 136 of Lecture Notes in Computer Science. Springer-Verlag Inc., New York, 1982. Hof89 C. Hoffman. The problem of accuracy and robustness in geometric computation. Computer, 22:3142, 1989. Hol75 J. H. Holland. Adaptation in Natural and Artificial Systems. University of Michigan Press, Ann Arbor, 1975. Hol81 I. Holyer. The NP-completeness of edge colorings. SIAM J. Computing, 10:718-720, 1981. Hol92 J. H. Holland. Genetic algorithms. Scientific American, 267(1):66-72, July 1992. Hop71 J. Hopcroft. An algorithm for minimizing the states in a finite automaton. In Z. Kohavi, editor, The theory of machines and computations, pages 189-196. Academic Press, New York, 1971. Hor80 R. N. Horspool. Practical fast searching in strings. Software - Practice and Experience, 10:501506, 1980. HP73 F. Harary and E. Palmer. Graphical enumeration. Academic Press, New York, 1973. HRW92 R. Hwang, D. Richards, and P. Winter. The Steiner Tree Problem, volume 53 of Annals of Discrete Mathematics. North Holland, Amsterdam, 1992. HS77 J. Hunt and T. Szymanski. A fast algorithm for computing longest common subsequences. Communications of the ACM, 20:350-353, 1977. HS94 J. Hershberger and J. Snoeyink. An implementation of the Douglas-Peucker algorithm for line simplification. In Proc. 10th Annu. ACM Sympos. Comput. Geom., pages 383-384, 1994.

file:///E|/BOOK/BOOK5/NODE232.HTM (24 of 50) [19/1/2003 1:28:02]

References

HSS87 J. Hopcroft, J. Schwartz, and M. Sharir. Planning, geometry, and complexity of robot motion. Ablex Publishing, Norwood NJ, 1987. HT73a J. Hopcroft and R. Tarjan. Dividing a graph into triconnected components. SIAM J. Computing, 2:135-158, 1973. HT73b J. Hopcroft and R. Tarjan. Efficient algorithms for graph manipulation. Communications of the ACM, 16:372-378, 1973. HT74 J. Hopcroft and R. Tarjan. Efficient planarity testing. J. ACM, 21:549-568, 1974. HT84 D. Harel and R. E. Tarjan. Fast algorithms for finding nearest common ancestors. SIAM J. Comput., 13:338-355, 1984. Hua92 X. Huang. A contig assembly program based on sensitive detection of fragment overlaps. Genomics, 1992. Hua94 X. Huang. On global sequence alignment. CABIOS, 10:227-235, 1994. Huf52 D. Huffman. A method for the construction of minimum-redundancy codes. Proc. of the IRE, 40:1098-1101, 1952. HW74 J. E. Hopcroft and J. K. Wong. Linear time algorithm for isomorphism of planar graphs. In Proc. Sixth Annual ACM Symposium on Theory of Computing, pages 172-184, 1974. HWK94 T. He, S. Wang, and A. Kaufman. Wavelet-based volume morphing. In Proc. IEEE Visualization '94, pages 85-92, 1994. HWL87 Jr. H. W. Lenstra. Factoring integers with elliptic curves. Annals of Mathematics, 126:649-673, 1987.

file:///E|/BOOK/BOOK5/NODE232.HTM (25 of 50) [19/1/2003 1:28:02]

References

IK75 O. Ibarra and C. Kim. Fast approximation algorithms for knapsack and sum of subset problems. J. ACM, 22:463-468, 1975. Ita78 A. Itai. Two commodity flow. J. ACM, 25:596-611, 1978. JAMS91 D. Johnson, C. Aragon, C. McGeoch, and D. Schevon. Optimization by simulated annealing: an experimental evaluation; partII, graph coloring and number partitioning. In Operations Research, volume 39, pages 378-406, 1991. Jan76 W. Janko. A list insertion sort for keys with arbitrary key distribution. ACM Trans. Math. Softw., 2(2):204-206, June 1976. Jar73 R. A. Jarvis. On the identification of the convex hull of a finite set of points in the plane. Info. Proc. Letters, 2:18-21, 1973. JD88 A. Jain and R. Dubes. Algorithms for Clustering Data. Prentice-Hall, Englewood Cliffs NJ, 1988. JM93 D. Johnson and C. McGeoch, editors. Network Flows and Matching: First DIMACS Implementation Challenge, volume 12. American Mathematics Society, Providence RI, 1993. Joh63 S. M. Johnson. Generation of permutations by adjacent transpositions. Math. Computation, 17:282285, 1963. Joh74 D. Johnson. Approximation algorithms for combinatorial problems. J. Computer and System Sciences, 9:256-278, 1974. Joh90 D. S. Johnson. A catalog of complexity classes. In J. van Leeuwen, editor, Handbook of Theoretical Computer Science: Algorithms and Complexity, volume A, pages 67-162. MIT Press, 1990. Jon86 D. W. Jones. An empirical comparison of priority-queue and event-set implementations. file:///E|/BOOK/BOOK5/NODE232.HTM (26 of 50) [19/1/2003 1:28:02]

References

Communications of the ACM, 29:300-311, 1986. Kah67 D. Kahn. The Code breakers: the story of secret writing. Macmillan, New York, 1967. Kah80 D. A. Kahamer. Fortran implementation of heap programs for efficient table maintenance. ACM Trans. Math. Softw., 6(3):444-449, September 1980. Kar72 R. M. Karp. Reducibility among combinatorial problems. In R. Miller and J. Thatcher, editors, Complexity of Computer Computations, pages 85-103. Plenum Press, 1972. Kar84 N. Karmarkar. A new polynomial-time algorithm for linear programming. Combinatorica, 4:373395, 1984. Kar96a D. Karger. Minimum cuts in near-linear time. In Proc. 28th ACM Symp. Theory of Computing, pages 56-63, 1996. Kar96b H. Karloff. How good is the Goemans-Williamson MAX CUT algorithm? In Proc. Twenty-Eighth Annual ACM Symposium on Theory of Computing, pages 427-434, 1996. Kei85 M. Keil. Decomposing a polygon into simpler components. SIAM J. Computing, 14:799-817, 1985. KGV83 S. Kirkpatrick, C. D. Gelatt, Jr., and M. P. Vecchi. Optimization by simulated annealing. Science, 220:671-680, 1983. KH80 J. Kennington and R. Helgason. Algorithms for Network Programming. Wiley-Interscience, New York, 1980. Kha79 L. Khachian. A polynomial algorithm in linear programming. Soviet Math. Dokl., 20:191-194, 1979. Kir79 file:///E|/BOOK/BOOK5/NODE232.HTM (27 of 50) [19/1/2003 1:28:02]

References

D. Kirkpatrick. Efficient computation of continuous skeletons. In Proc. 20th IEEE Symp. Foundations of Computing, pages 28-35, 1979. Kir83 D. Kirkpatrick. Optimal search in planar subdivisions. SIAM J. Computing, 12:28-35, 1983. KKT95 D. Karger, P. Klein, and R. Tarjan. A randomized linear-time algorithm to find minimum spanning trees. J. ACM, 42:321-328, 1995. KL70 B. W. Kernighan and S. Lin. An efficient heuristic procedure for partitioning graphs. The Bell System Technical Journal, pages 291-307, 1970. KM72 V. Klee and G. Minty. How good is the simplex algorithm. In Inequalities III, pages 159-172, New York, 1972. Academic Press. KM95 J. D. Kececioglu and E. W. Myers. Combinatorial algorithms for DNA sequence assembly. Algorithmica, 13(1/2):7-51, January 1995. KMP77 D. Knuth, J. Morris, and V. Pratt. Fast pattern matching in strings. SIAM J. Computing, 6:323-350, 1977. Kno79 H. D. Knoble. An efficient one-way enciphering algorithm. ACM Trans. Math. Softw., 5(1):108111, March 1979. Knu73a D. E. Knuth. The Art of Computer Programming, Volume 1: Fundamental Algorithms. AddisonWesley, Reading MA, second edition, 1973. Knu73b D. E. Knuth. The Art of Computer Programming, Volume 3: Sorting and Searching. AddisonWesley, Reading MA, 1973. Knu81 D. E. Knuth. The Art of Computer Programming, Volume 2: Seminumerical Algorithms. AddisonWesley, Reading MA, second edition, 1981.

file:///E|/BOOK/BOOK5/NODE232.HTM (28 of 50) [19/1/2003 1:28:02]

References

Knu94 D. Knuth. The Stanford GraphBase: a platform for combinatorial computing. ACM Press, New York, 1994. KO63 A. Karatsuba and Yu. Ofman. Multiplication of multi-digit numbers on automata. Sov. Phys. Dokl., 7:595-596, 1963. KOS91 A. Kaul, M. A. O'Connor, and V. Srinivasan. Computing Minkowski sums of regular polygons. In Proc. 3rd Canad. Conf. Comput. Geom., pages 74-77, 1991. Koz92 J. R. Koza. Genetic Programming. MIT Press, Cambridge, MA, 1992. KPS89 B. Korte, H. Prömel, and A. Steger. Steiner trees in VLSI-layout. In B. Korte, L. Lovasz, H. Prömel, and A. Schrijver, editors, Paths, Flows, and VLSI-layout, pages 185-214. SpringerVerlag, Berlin, 1989. KR87 R. Karp and M. Rabin. Efficient randomized pattern-matching algorithms. IBM J. Research and Development, 31:249-260, 1987. KR91 A. Kanevsky and V. Ramachandran. Improved algorithms for graph four-connectivity. J. Comp. Sys. Sci., 42:288-306, 1991. Kru56 J. B. Kruskal. On the shortest spanning subtree of a graph and the traveling salesman problem. Proc. of the American Mathematical Society, 7:48-50, 1956. KS85 M. Keil and J. R. Sack. Computational Geometry, chapter Minimum decomposition of geometric objects, pages 197-216. North-Holland, 1985. KS86 D. Kirkpatrick and R. Siedel. The ultimate planar convex hull algorithm? SIAM J. Computing, 15:287-299, 1986. KS90 K. Kedem and M. Sharir. An efficient motion planning algorithm for a convex rigid polygonal file:///E|/BOOK/BOOK5/NODE232.HTM (29 of 50) [19/1/2003 1:28:02]

References

object in 2-dimensional polygonal space. Discrete and Computational Geometry, 5:43-75, 1990. Kuh75 H. W. Kuhn. Steiner's problem revisited. In G. Dantzig and B. Eaves, editors, Studies in Optimization, pages 53-70. Mathematical Association of America, 1975. Kur30 K. Kuratowski. Sur le problème des courbes gauches en topologie. Fund. Math., 15:217-283, 1930. Kwa62 M. Kwan. Graphic programming using odd and even points. Chinese Math., 1:273-277, 1962. Lat91 J.-C. Latombe. Robot Motion Planning. Kluwer Academic Publishers, Boston, 1991. Law76 E. Lawler. Combinatorial Optimization: Networks and Matroids. Holt, Rinehart, and Winston, Fort Worth TX, 1976. Lee82 D. T. Lee. Medial axis transformation of a planar shape. IEEE Trans. Pattern Analysis and Machine Intelligence, PAMI-4:363-369, 1982. Lee94 H. Leeb. pLab - a system for testing random numbers. In Proc. Int. Workshop Parallel Numerics, pages 89-99. Slovak Academy of Science, Institute for Informatics, Slovakia, 1994. Also available via ftp from http://random.mat.sbg.ac.at/ftp/pub/publications/leeb/smolenice/. Len87 T. Lengauer. Efficient algorithms for finding minimum spanning forests of hierarchically defined graphs. J. Algorithms, 8, 1987. Len89 T. Lengauer. Hierarchical planarity testing algorithms. J. ACM, 36(3):474-509, July 1989. Len90 T. Lengauer. Combinatorial Algorithms for Integrated Circuit Layout. Wiley, Chichester, England, 1990. Lev92 J. L. Leva. A normal random number generator. ACM Trans. Math. Softw., 18(4):454-455, December 1992. file:///E|/BOOK/BOOK5/NODE232.HTM (30 of 50) [19/1/2003 1:28:02]

References

Lew82 J. G. Lewis. The Gibbs-Poole-Stockmeyer and Gibbs-King algorithms for reordering sparse matrices. ACM Trans. Math. Softw., 8(2):190-194, June 1982. Lin65 S. Lin. Computer solutions of the traveling salesman problem. Bell System Tech. J., 44:2245-2269, 1965. LK73 S. Lin and B. Kernighan. An effective heuristic algorithm for the traveling salesman problem. Operations Research, 21:498-516, 1973. LLK83 J. K. Lenstra, E. L. Lawler, and A. Rinnooy Kan. Theory of Sequencing and Scheduling. Wiley, New York, 1983. LLKS85 E. Lawler, J. Lenstra, A. Rinnooy Kan, and D. Shmoys. The Traveling Salesman Problem. John Wiley, 1985. LLS92 L. Lam, S.-W. Lee, and C. Suen. Thinning methodologies - a comprehensive survey. IEEE Trans. Pattern Analysis and Machine Intelligence, 14:869-885, 1992. LP86 L. Lovász and M. Plummer. Matching Theory. North-Holland, Amsterdam, 1986. LPW79 T. Lozano-Perez and M. Wesley. An algorithm for planning collision-free paths among polygonal obstacles. Comm. ACM, 22:560-570, 1979. LR93 K. Lang and S. Rao. Finding near-optimal cuts: An empirical evaluation. In Proc. 4th Annual ACMSIAM Symposium on Discrete Algorithms (SODA '93), pages 212-221, 1993. LS87 V. Lumelski and A. Stepanov. Path planning strategies for a point mobile automaton moving amidst unknown obstacles of arbitrary shape. Algorithmica, 3:403-430, 1987. LS95 Y.-L. Lin and S. Skiena. Algorithms for square roots of graphs. SIAM J. Discrete Mathematics,

file:///E|/BOOK/BOOK5/NODE232.HTM (31 of 50) [19/1/2003 1:28:02]

References

8:99-118, 1995. LT79 R. Lipton and R. Tarjan. A separator theorem for planar graphs. SIAM Journal on Applied Mathematics, 36:346-358, 1979. LT80 R. Lipton and R. Tarjan. Applications of a planar separator theorem. SIAM J. Computing, 9:615626, 1980. Luc91 E. Lucas. Récréations Mathématiques. Gauthier-Villares, Paris, 1891. Luk80 E. M. Luks. Isomorphism of bounded valence can be tested in polynomial time. In Proc. of the 21st Annual Symposium on Foundations of Computing, pages 42-49. IEEE, 1980. LV93 M. Li and P. Vitáni. An introduction to Kolmogorov complexity and its applications. SpringerVerlag, New York, 1993. LW77 D. T. Lee and C. K. Wong. Worst-case analysis for region and partial region searches in multidimensional binary search trees and balanced quad trees. Acta Informatica, 9:23-29, 1977. LW88 T. Lengauer and E. Wanke. Efficient solution of connectivity problems on hierarchically defined graphs. SIAM J. Computing, 17:1063-1080, 1988. LY93 C. Lund and M. Yannakakis. On the hardness of approximating minimization problems. In Proc. 25th ACM Symp. Theory of Computing (STOC), pages 286-293, 1993. Man89 U. Manber. Introduction to Algorithms. Addison-Wesley, Reading MA, 1989. Mar83 S. Martello. An enumerative algorithm for finding Hamiltonian circuits in a directed graph. ACM Trans. Math. Softw., 9(1):131-138, March 1983. Mat87 D. W. Matula. Determining edge connectivity in O(nm). In 28th Ann. Symp. Foundations of file:///E|/BOOK/BOOK5/NODE232.HTM (32 of 50) [19/1/2003 1:28:02]

References

Computer Science, pages 249-251. IEEE, 1987. McC76 E. McCreight. A space-economical suffix tree construction algorithm. J. ACM, 23:262-272, 1976. McH90 J. McHugh. Algorithmic Graph Theory. Prentice-Hall, Englewood Cliffs NJ, 1990. McN83 J. M. McNamee. A sparse matrix package - part II: Special cases. ACM Trans. Math. Softw., 9(3):344-345, September 1983. Meg83 N. Megiddo. Linear time algorithm for linear programming in Computing, 12:759-776, 1983.

and related problems. SIAM J.

Meh84 K. Mehlhorn. Data structures and algorithms, volume 1-3. Springer-Verlag, Berlin, 1984. Men27 K. Menger. Zur allgemeinen kurventheorie. Fund. Math., 10:96-115, 1927. MGH81 J. J. Moré, B. S. Garbow, and K. E. Hillstrom. Fortran subroutines for testing unconstrained optimization software. ACM Trans. Math. Softw., 7(1):136-140, March 1981. MH78 R. Merkle and M Hellman. Hiding and signatures in trapdoor knapsacks. IEEE Trans. Information Theory, 24:525-530, 1978. Mic92 Z. Michalewicz. Genetic Algorithms + Data Structures = Evolution Programs. Springer, Berlin, 1992. Mie58 W. Miehle. Link-minimization in networks. Operations Research, 6:232-243, 1958. Mil76 G. Miller. Reimann's hypothesis and tests for primality. J. Computer and System Sciences, 13:300317, 1976. Mil89 file:///E|/BOOK/BOOK5/NODE232.HTM (33 of 50) [19/1/2003 1:28:02]

References

V. Milenkovic. Double precision geometry: a general technique for calculating line and segment intersections using rounded arithmetic. In Proc. 30th IEEE Symp. Foundations of Computer Science (FOCS), pages 500-505, 1989. Mil97 V. Milenkovic. Multiple translational containment. part II: exact algorithms. Algorithmica, 1997. Min78 H. Minc. Permanents, volume 6 of Encyclopedia of Mathematics and its Applications. AddisonWesley, Reading MA, 1978. Mit96 J. Mitchell. Guillotine subdivisions approximate polygonal subdivisions: Part II - A simple polynomial-time approximation scheme for geometric k-MST, TSP, and related problems. University at Stony Brook, Part I appears in SODA'96, pp. 402-408, 1996. MM90 U. Manber and G. Myers. Suffix arrays: A new method for on-line string searches. In Proceedings First ACM-SIAM Symposium on Discrete Algorithms, pages 319-327. SIAM, 1990. MMI72 D. Matula, G. Marble, and J. Isaacson. Graph coloring algorithms. In R. C. Read, editor, Graph Theory and Computing, pages 109-122. Academic Press, 1972. MN95 K. Mehlhorn and S. Näher. LEDA, a platform for combinatorial and geometric computing. Communications of the ACM, 38:96-102, 1995. MO63 L. E. Moses and R. V. Oakford. Tables of Random Permutations. Stanford University Press, Stanford, Calif., 1963. Moo59 E. F. Moore. The shortest path in a maze. In Proc. International Symp. Switching Theory, pages 285-292. Harvard University Press, 1959. MP78 D. Muller and F. Preparata. Finding the intersection of two convex polyhedra. Theoretical Computer Science, 7:217-236, 1978. MP80 W. Masek and M. Paterson. A faster algorithm for computing string edit distances. J. Computer file:///E|/BOOK/BOOK5/NODE232.HTM (34 of 50) [19/1/2003 1:28:02]

References

and System Sciences, 20:18-31, 1980. MR95 R. Motwani and P. Raghavan. Randomized Algorithms. Cambridge University Press, New York, 1995. MRRT53 N. Metropolis, A. W. Rosenbluth, M. N. Rosenbluth, and A. H. Teller. Equation of state calculations by fast computing machines. Journal of Chemical Physics, 21(6):1087-1092, June 1953. MS91 B. Moret and H. Shapiro. Algorithm from P to NP: Design and Efficiency. Benjamin/Cummings, Redwood City, CA, 1991. MS93 M. Murphy and S. Skiena. Ranger: A tool for nearest neighbor search in high dimensions. In Proc. Ninth ACM Symposium on Computational Geometry, pages 403-404, 1993. MS95a D. Margaritis and S. Skiena. Reconstructing strings from substrings in rounds. Proc. 36th IEEE Symp. Foundations of Computer Science (FOCS), 1995. MS95b J. S. B. Mitchell and S. Suri. Separation and approximation of polyhedral objects. Comput. Geom. Theory Appl., 5:95-114, 1995. MT85 S. Martello and P. Toth. A program for the 0-1 multiple knapsack problem. ACM Trans. Math. Softw., 11(2):135-140, June 1985. MT87 S. Martello and P. Toth. Algorithms for knapsack problems. In S. Martello, editor, Surveys in Combinatorial Optimization, volume 31 of Annals of Discrete Mathematics, pages 213-258. NorthHolland, 1987. MT90a S. Martello and P. Toth. Knapsack problems: algorithms and computer implementations. Wiley, New York, 1990. MT90b K. Mehlhorn and A. Tsakalidis. Data structures. In J. van Leeuwen, editor, Handbook of file:///E|/BOOK/BOOK5/NODE232.HTM (35 of 50) [19/1/2003 1:28:02]

References

Theoretical Computer Science: Algorithms and Complexity, volume A, pages 301-341. MIT Press, 1990. Mul94 K. Mulmuley. Computational Geometry: an introduction through randomized algorithms. PrenticeHall, New York, 1994. MV80 S. Micali and V. Vazirani. An O

algorithm for finding maximum matchings in general

graphs. In Proc. 21st. Symp. Foundations of Computing, pages 17-27, 1980. MW93 J. More and S. Wright. Optimization Software Guide. SIAM, Philadelphia PA, 1993. Mye86 E. Myers. An O(n d) difference algorithm and its variations. Algorithmica, 1:514-534, 1986. NC88 T. Nishizeki and N. Chiba. Planar Graphs: Theory and Algorithms. North-Holland, Amsterdam, 1988. NH93 J. Nievergelt and K. Hinrichs. Algorithms and Data Structures with applications to graphics and geometry. Prentice Hall, Englewood Cliffs NJ, 1993. NU95 Stefan Näher and C. Uhrig. The LEDA user manual, version r3.2. Available by ftp from ftp.mpisb.mpg.de in directory /pub/LEDA, 1995. NW78 A. Nijenhuis and H. Wilf. Combinatorial Algorithms for Computers and Calculators. Academic Press, Orlando FL, second edition, 1978. NW86 J. C. Nash and R. L. C. Wang. Subroutines for testing programs that compute the generalized inverse of a matrix. ACM Trans. Math. Softw., 12(3):274-277, September 1986. NZ80 I. Niven and H. Zuckerman. An Introduction to the Theory of Numbers. Wiley, New York, fourth edition, 1980.

file:///E|/BOOK/BOOK5/NODE232.HTM (36 of 50) [19/1/2003 1:28:02]

References

Ogn93 R. Ogniewicz. Discrete Voronoi Skeletons. Hartung-Gorre Verlag, Konstanz, Germany, 1993. O'R87 J. O'Rourke. Art Gallery Theorems and Algorithms. Oxford University Press, Oxford, 1987. O'R94 J. O'Rourke. Computational Geometry in C. Cambridge University Press, New York, 1994. Ort88 J. Ortega. Introduction to Parallel and Vector Solution of Linear Systems. Plenum, New York, 1988. OvL81 M. Overmars and J. van Leeuwen. Maintenance of configurations in the plane. J. Computer and System Sciences, 23:166-204, 1981. OW85 J. O'Rourke and R. Washington. Curve similarity via signatures. In G. T. Toussaint, editor, Computational Geometry, pages 295-317. North-Holland, Amsterdam, Netherlands, 1985. Pag74 R. L. Page. A minimal spanning tree clustering method. Commun. ACM, 17(6):321-323, June 1974. Pal85 E. M. Palmer. Graphical Evolution: An Introduction to the Theory of Random Graphs. WileyInterscience, New York, 1985. Pap76a C. Papadimitriou. The complexity of edge traversing. J. ACM, 23:544-554, 1976. Pap76b C. Papadimitriou. The NP-completeness of the bandwidth minimization problem. Computing, 16:263-270, 1976. Pap80 U. Pape. Shortest path lengths. ACM Trans. Math. Softw., 6(3):450-455, September 1980. Par90 G. Parker. A better phonetic search. C Gazette, 5-4, June/July 1990.

file:///E|/BOOK/BOOK5/NODE232.HTM (37 of 50) [19/1/2003 1:28:02]

References

Pav82 T. Pavlidis. Algorithms for Graphics and Image Processing. Computer Science Press, Rockville MD, 1982. PFTV86 W. Press, B. Flannery, S. Teukolsky, and W. T. Vetterling. Numerical Recipes: the art of scientific computing. Cambridge University Press, 1986. PGD82 D. Phillips and A. Garcia-Diaz. Fundamentals of Network Analysis. Prentice-Hall, Englewood Cliffs NJ, 1982. PH80 M. Padberg and S. Hong. On the symmetric traveling salesman problem: a computational study. Math. Programming Studies, 12:78-107, 1980. PL94 P. A. Pevzner and R. J. Lipshutz. Towards DNA sequencing chips. In 19th Int. Conf. Mathematical Foundations of Computer Science, volume 841, pages 143-158, Lecture Notes in Computer Science, 1994. PM88 S. Park and K. Miller. Random number generators: Good ones are hard to find. Communications of the ACM, 31:1192-1201, 1988. Pol57 G. Polya. How to Solve it. Princeton University Press, Princeton NJ, second edition, 1957. Pom84 C. Pomerance. The quadratic sieve factoring algorithm. In T. Beth, N. Cot, and I. Ingemarrson, editors, Advances in Cryptology, volume 209, pages 169-182. Lecture Notes in Computer Science, Springer-Verlag, 1984. PR79 S. Peleg and A. Rosenfield. Breaking substitution ciphers using a relaxation algorithm. Comm. ACM, 22:598-605, 1979. Pra75 V. Pratt. Every prime has a succinct certificate. SIAM J. Computing, 4:214-220, 1975. Pri57 R. C. Prim. Shortest connection networks and some generalizations. Bell System Technical file:///E|/BOOK/BOOK5/NODE232.HTM (38 of 50) [19/1/2003 1:28:02]

References

Journal, 36:1389-1401, 1957. Prü18 H. Prüfer. Neuer beweis eines satzes über permutationen. Arch. Math. Phys., 27:742-744, 1918. PS82 C. Papadimitriou and K. Steiglitz. Combinatorial Optimization: Algorithms and Complexity. Prentice-Hall, Englewood Cliffs, NJ, 1982. PS85 F. Preparata and M. Shamos. Computational Geometry. Springer-Verlag, New York, 1985. PSW92 T. Pavlides, J. Swartz, and Y. Wang. Information encoding with two-dimensional bar-codes. IEEE Computer, 25:18-28, 1992. Pug90 W. Pugh. Skip lists: A probabilistic alternative to balanced trees. Communications of the ACM, 33:668-676, 1990. PW83 S. Pizer and V. Wallace. To compute numerically: concepts and strategies. Little, Brown, Boston, 1983. Rab80 M. Rabin. Probabilistic algorithm for testing primality. J. Number Theory, 12:128-138, 1980. Rab95 F. M. Rabinowitz. A stochastic algorithm for global optimization with constraints. ACM Trans. Math. Softw., 21(2):194-213, June 1995. Raw92 G. Rawlins. Compared to What? Computer Science Press, New York, 1992. RC55 Rand-Corporation. A million random digits with 100,000 normal deviates. The Free Press, Glencoe, IL, 1955. RDC93 E. Reingold, N. Dershowitz, and S. Clamen. Calendrical calculations II: Three historical calendars. Software - Practice and Experience, 22:383-404, 1993.

file:///E|/BOOK/BOOK5/NODE232.HTM (39 of 50) [19/1/2003 1:28:02]

References

Rei72 E. Reingold. On the optimality of some set algorithms. J. ACM, 19:649-659, 1972. Rei91 G. Reinelt. TSPLIB - a traveling salesman problem library. ORSA J. Computing, 3:376-384, 1991. Rei94 G. Reinelt. The traveling salesman problem: Computational solutions for TSP applications. In Lecture Notes in Computer Science 840, pages 172-186. Springer-Verlag, Berlin, 1994. Ren84 R. J. Renka. Triangulation and interpolation at arbitrarily distributed points in the plane. ACM Trans. Math. Softw., 10(4):440-442, December 1984. RHS89 A. Robison, B. Hafner, and S. Skiena. Eight pieces cannot cover a chessboard. Computer Journal, 32:567-570, 1989. Riv77 R. Rivest. On the worst-case behavior of string-searching algorithms. SIAM J. Computing, 6:669674, 1977. Riv92 R. Rivest. The MD5 message digest algorithm. RFC 1321, 1992. RND77 E. Reingold, J. Nievergelt, and N. Deo. Combinatorial algorithms: theory and practice. PrenticeHall, Englewood Cliffs NJ, 1977. RS96 H. Rau and S. Skiena. Dialing for documents: an experiment in information theory. Journal of Visual Languages and Computing, pages 79-95, 1996. RSA78 R. Rivest, A. Shamir, and L. Adleman. A method for obtaining digital signatures and public-key cryptosystems. Communications of the ACM, 21:120-126, 1978. RSL77 D. Rosenkrantz, R. Stearns, and P. M. Lewis. An analysis of several heuristics for the traveling salesman problem. SIAM J. Computing, 6:563-581, 1977. RSST96 file:///E|/BOOK/BOOK5/NODE232.HTM (40 of 50) [19/1/2003 1:28:02]

References

N. Robertson, D. Sanders, P. Seymour, and R. Thomas. Efficiently four-coloring planar graphs. In Proc. 28th ACM Symp. Theory of Computing, pages 571-575, 1996. RT81 E. Reingold and J. Tilford. Tidier drawings of trees. IEEE Trans. Software Engineering, 7:223228, 1981. Rus97 F. Ruskey. Combinatorial generation. book in preparation, 1997. Ryt85 W. Rytter. Fast recognition of pushdown automaton and context-free languages. Information and Control, 67:12-22, 1985. SA95 M. Sharir and P. Agarwal. Davenport-Schinzel sequences and their geometric applications. Cambridge University Press, New York, 1995. Sam90a H. Samet. Applications of spatial data structures. Addison-Wesley, Reading MA, 1990. Sam90b H. Samet. The design and analysis of spatial data structures. Addison-Wesley, Reading MA, 1990. Sax80 J. B. Saxe. Dynamic programming algorithms for recognizing small-bandwidth graphs in polynomial time. SIAM J. Algebraic and Discrete Methods, 1:363-369, 1980. Sch94 B. Schneier. Applied Cryptography. Wiley, New York, 1994. SD75 M. Syslo and J. Dzikiewicz. Computational experiences with some transitive closure algorithms. Computing, 15:33-39, 1975. SD76 D. C. Schmidt and L. E. Druffel. A fast backtracking algorithm to test directed graphs for isomorphism using distance matrices. J. ACM, 23:433-445, 1976. SDK83 M. Syslo, N. Deo, and J. Kowalik. Discrete Optimization Algorithms with Pascal Programs. file:///E|/BOOK/BOOK5/NODE232.HTM (41 of 50) [19/1/2003 1:28:02]

References

Prentice Hall, Englewood Cliffs NJ, 1983. Sed77 R. Sedgewick. Permutation generation methods. Computing Surveys, 9:137-164, 1977. Sed78 R. Sedgewick. Implementing quicksort programs. Communications of the ACM, 21:847-857, 1978. Sed92 R. Sedgewick. Algorithms in C++. Addison-Wesley, Reading MA, 1992. SF92 T. Schlick and A. Fogelson. TNPACK - a truncated Newton minimization package for large-scale problems: I. algorithm and usage. ACM Trans. Math. Softw., 18(1):46-70, March 1992. SFG82 M. Shore, L. Foulds, and P. Gibbons. An algorithms for the Steiner problem in graphs. Networks, 12:323-333, 1982. SH75 M. Shamos and D. Hoey. Closest point problems. In Proc. Sixteenth IEEE Symp. Foundations of Computer Science, pages 151-162, 1975. SH76 M. Shamos and D. Hoey. Geometric intersection problems. In Proc. 17th IEEE Symp. Foundations of Computer Science, pages 208-215, 1976. Sha78 M. Shamos. Computational Geometry. PhD thesis, Yale University, UMI #7819047, 1978. Sha87 M. Sharir. Efficient algorithms for planning purely translational collision-free motion in two and three dimensions. In Proc. IEEE Internat. Conf. Robot. Autom., pages 1326-1331, 1987. Sha93 R. Sharda. Linear and Discrete Optimization and Modeling Software: A Resource Handbook. Lionheart Publishing Inc., 1993. She78 A. H. Sherman. NSPIV: a Fortran subroutine for sparse Gaussian elimination with partial pivoting. ACM Trans. Math. Softw., 4(4):391-398, December 1978.

file:///E|/BOOK/BOOK5/NODE232.HTM (42 of 50) [19/1/2003 1:28:02]

References

She96 J. R. Shewchuk. Robust adaptive floating-point geometric predicates. In Proc. 12th ACM Computational Geometry, pages 141-150, 1996. SK83 D. Sankoff and J. Kruskal. Time Warps, String Edits, and Macromolecules: the theory and practice of sequence comparison. Addison-Wesley, Reading MA, 1983. SK86 T. Saaty and P. Kainen. The Four-Color Problem. Dover, New York, 1986. SK93 R. Skeel and J. Keiper. Elementary Numerical computing with Mathematica. McGraw-Hill, New York, 1993. Ski88 S. Skiena. Encroaching lists as a measure of presortedness. BIT, 28:775-784, 1988. Ski90 S. Skiena. Implementing Discrete Mathematics. Addison-Wesley, Redwood City, CA, 1990. Sla61 P. Slater. Inconsistencies in a schedule of paired comparisons. Biometrika, 48:303-312, 1961. SM73 L. Stockmeyer and A. Meyer. Word problems requiring exponential time. In Proc. Fifth ACM Symp. Theory of Computing, pages 1-9, 1973. Smi91 D. M. Smith. A Fortran package for floating-point multiple-precision arithmetic. ACM Trans. Math. Softw., 17(2):273-283, June 1991. Sou96 E. Southern. DNA chips: analysing sequence by hybridization to oligonucleotides on a large scale. Trends in Genetics, 12:110-115, 1996. SPP76 A. Schönhage, M. Paterson, and N. Pippenger. Finding the median. J. Computer and System Sciences, 13:184-199, 1976. SR83 K. Supowit and E. Reingold. The complexity of drawing trees nicely. Acta Informatica, 18:377file:///E|/BOOK/BOOK5/NODE232.HTM (43 of 50) [19/1/2003 1:28:02]

References

392, 1983. SR95 R. Sharda and G. Rampal. Algebraic modeling languages on PCs. OR/MS Today, 22-3:58-63, 1995. SS71 A. Schönhage and V. Strassen. Schnele multiplikation grosser zahlen. Computing, 7:281-292, 1971. SS90 J. T. Schwartz and M. Sharir. Algorithmic motion planning. In J. van Leeuwen, editor, Handbook of Theoretical Computer Science: Algorithms and Complexity, volume A, pages 391-430. MIT Press, 1990. SSS74 I. Sutherland, R. Sproull, and R. Shumacker. A characterization of ten hidden surface algorithms. ACM Computing Surveys, 6:1-55, 1974. ST85 D. Sleator and R. Tarjan. Self-adjusting binary search trees. J. ACM, 32:652-686, 1985. Sta90 J. Stasko. Tango: A framework and system for algorithm animation. Computer, 23:27-39, 1990. Sta95 W. Stallings. Protect your Privacy: a guide for PGP Users. Prentice Hall PTR, Englewood Cliffs NJ, 1995. Ste94 G. A. Stephen. String Searching Algorithms. Lecture-Notes-Series-on-Computing. WorldScientific-Publishing, October 1994. Sto88 J. Storer. Data compression: methods and theory. Computer Science Press, Rockville MD, 1988. Str69 V. Strassen. Gaussian elimination is not optimal. Numerische Mathematik, 14:354-356, 1969. SV87 J. Stasko and J. Vitter. Pairing heaps: Experiments and analysis. Communications of the ACM, 30(3):234-249, 1987. file:///E|/BOOK/BOOK5/NODE232.HTM (44 of 50) [19/1/2003 1:28:02]

References

SV88 B. Schieber and U. Vishkin. On finding lowest common ancestors: simplification and parallelization. SIAM J. Comput., 17(6):1253-1262, December 1988. SW86 D. Stanton and D. White. Constructive Combinatorics. Springer-Verlag, New York, 1986. SW94 M. Stoer and F. Wagner. A simple min cut algorithm. In European Symp. Algorithms (ESA), volume 855, pages 141-147. Springer Verlag Lecture Notes in Computer Science, 1994. SW95 J. Salowe and D. Warme. Thirty-five-point rectilinear Steiner minimal trees in a day. Networks: An International Journal, 25, 1995. SWM95 J. Shallit, H. Williams, and F. Moraine. Discovery of a lost factoring machine. The Mathematical Intelligencer, 17-3:41-47, Summer 1995. Tar95 G. Tarry. Le problème de labyrinthes. Nouvelles Ann. de Math., 14:187, 1895. Tar72 R. Tarjan. Depth-first search and linear graph algorithms. SIAM J. Computing, 1:146-160, 1972. Tar75 R. Tarjan. Efficiency of a good by not linear set union algorithm. J. ACM, 22:215-225, 1975. Tar79 R. Tarjan. A class of algorithms which require non-linear time to maintain disjoint sets. J. Computer and System Sciences, 18:110-127, 1979. Tar83 R. Tarjan. Data Structures and Network Algorithms. Society for Industrial and Applied Mathematics, Philadelphia, 1983. Tho68 K. Thompson. Regular expression search algorithm. Communications of the ACM, 11:419-422, 1968. Tho96 file:///E|/BOOK/BOOK5/NODE232.HTM (45 of 50) [19/1/2003 1:28:02]

References

Mikkel Thorup. On RAM priority queues. In Proc. Seventh Annual ACM-SIAM Symposium on Discrete Algorithms, pages 59-67, 1996. Tin90 G. Tinhofer. Generating graphs uniformly at random. Computing, 7:235-255, 1990. Tro62 H. F. Trotter. Perm (algorithm 115). Comm. ACM, 5:434-435, 1962. Tur88 J. Turner. Almost all k-colorable graphs are easy to color. J. Algorithms, 9:63-82, 1988. TW88 R. Tarjan and C. Van Wyk. An O Computing, 17:143-178, 1988.

algorithm for triangulating a simple polygon. SIAM J.

Ukk92 E. Ukkonen. Constructing suffix trees on-line in linear time. In Intern. Federation of Information Processing (IFIP '92), pages 484-492, 1992. Vai88 P. Vaidya. Geometry helps in matching. In Proc. 20th ACM Symp. Theory of Computing, pages 422-425, 1988. Val79 L. Valiant. The complexity of computing the permanent. Theoretical Computer Science, 8:189201, 1979. Var91 I. Vardi. Computational Recreations in Mathematica. Addison-Wesley, Redwood City CA, 1991. Vau80 J. Vaucher. Pretty printing of trees. Software Practice and Experience, 10:553-561, 1980. vEBKZ77 P. van Emde Boas, R. Kaas, and E. Zulstra. Design and implementation of an efficient priority queue. Math. Systems Theory, 10:99-127, 1977. Veg90 G. Vegter. The visibility diagram: a data structure for visibility problems and motion planning. In Proc. second Scand. Workshop on Algorithm Theory (SWAT), volume 447, pages 97-110. Springer Verlag Lecture Notes in Computer Science, 1990. file:///E|/BOOK/BOOK5/NODE232.HTM (46 of 50) [19/1/2003 1:28:02]

References

Vit89 J. S. Vitter. Dynamic Huffman coding. ACM Trans. Math. Softw., 15(2):158-167, June 1989. Viz64 V. G. Vizing. On an estimate of the chromatic class of a p-graph (in Russian). Diskret. Analiz, 3:2330, 1964. vL90a J. van Leeuwen. Graph algorithms. In J. van Leeuwen, editor, Handbook of Theoretical Computer Science: Algorithms and Complexity, volume A, pages 525-631. MIT Press, 1990. vL90b J. van Leeuwen, editor. Handbook of Theoretical Computer Science: Algorithms and Complexity, volume A. MIT Press, 1990. vN63 J. von Neumann. Various techniques used in connection with random digits. In A. H. Traub, editor, John von Neumann, Collected Works, volume 5. Macmillan, 1963. Vos92 S. Voss. Steiner's problem in graphs: heuristic methods. Discrete Applied Mathematics, 40, 1992. War62 S. Warshall. A theorem on boolean matrices. J. ACM, 9:11-12, 1962. Wat95 M. Waterman. Introduction to Computational Biology. Chapman Hall, London, 1995. WBCS77 J. Weglarz, J. Blazewicz, W. Cellary, and R. Slowinski. An automatic revised simplex method for constrained resource network scheduling. ACM Trans. Math. Softw., 3(3):295-300, September 1977. Wei73 P. Weiner. Linear pattern-matching algorithms. In Proc. 14th IEEE Symp. on Switching and Automata Theory, pages 1-11, 1973. Wei92 M. Weiss. Data Structures and Algorithm Analysis. Benjamin/Cummings, Redwood City, CA, 1992.

file:///E|/BOOK/BOOK5/NODE232.HTM (47 of 50) [19/1/2003 1:28:02]

References

Wel84 T. Welch. A technique for high-performance data compression. IEEE Computer, 17-6:8-19, 1984. Wes83 D. H. West. Approximate solution of the quadratic assignment problem. ACM Trans. Math. Softw., 9(4):461-466, December 1983. WF74 R. A. Wagner and M. J. Fischer. The string-to-string correction problem. J. ACM, 21:168-173, 1974. Whi32 H. Whitney. Congruent graphs and the connectivity of graphs. American J. Mathematics, 54:150168, 1932. Wig83 A. Wigerson. Improving the performance guarantee for approximate graph coloring. J. ACM, 30:729-735, 1983. Wil64 J. W. J. Williams. Algorithm 232 (heapsort). Communications of the ACM, 7:347-348, 1964. Wil84 H. Wilf. Backtrack: An O(1) expected time algorithm for graph coloring. Info. Proc. Letters, 18:119-121, 1984. Wil85 D. E. Willard. New data structures for orthogonal range queries. SIAM J. Computing, 14:232-253, 1985. Wil89 H. Wilf. Combinatorial Algorithms: an update. SIAM, Philadelphia PA, 1989. Win68 S. Winograd. A new algorithm for inner product. IEEE Trans. Computers, C-17:693-694, 1968. Win80 S. Winograd. Arithmetic Complexity of Computations. SIAM, Philadelphia, 1980. WM92a S. Wu and U. Manber. Agrep - a fast approximate pattern-matching tool. In Usenix Winter 1992 Technical Conference, pages 153-162, 1992. file:///E|/BOOK/BOOK5/NODE232.HTM (48 of 50) [19/1/2003 1:28:02]

References

WM92b S. Wu and U. Manber. Fast text searching allowing errors. Comm. ACM, 35:83-91, 1992. Wol79 T. Wolfe. The Right Stuff. Bantam Books, Toronto, 1979. Woo93 D. Wood. Data Structures, Algorithms, and Performance. Addison-Wesley, Reading MA, 1993. WS79 C. Wetherell and A. Shannon. Tidy drawing of trees. IEEE Trans. Software Engineering, 5:514520, 1979. WW95 F. Wagner and A. Wolff. Map labeling heuristics: provably good and practically useful. In Proc. 11th ACM Symp. Computational Geometry, pages 109-118, 1995. Yao81 A. C. Yao. A lower bound to finding convex hulls. J. ACM, 28:780-787, 1981. Yao90 F. Yao. Computational geometry. In J. van Leeuwen, editor, Handbook of Theoretical Computer Science: Algorithms and Complexity, volume A, pages 343-389. MIT Press, 1990. YS96 F. Younas and S. Skiena. Randomized algorithms for identifying minimal lottery ticket sets. Journal of Undergraduate Research, 2-2:88-97, 1996. YY76 A. Yao and F. Yao. The complexity of searching in an ordered random table. In Proc. 17th IEEE Symp. Foundations of Computer Science, pages 173-177, 1976. ZL78 J. Ziv and A. Lempel. A universal algorithm for sequential data compression. IEEE Trans. Information Theory, IT-23:337-343, 1978.

file:///E|/BOOK/BOOK5/NODE232.HTM (49 of 50) [19/1/2003 1:28:02]

References

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE232.HTM (50 of 50) [19/1/2003 1:28:02]

About this document ...

Up: The Algorithm Design Manual Previous: Index

About this document ... The Algorithm Design Manual This document was generated using the LaTeX2HTML translator Version 96.1 (Feb 5, 1996) Copyright © 1993, 1994, 1995, 1996, Nikos Drakos, Computer Based Learning Unit, University of Leeds. The command line arguments were: latex2html book.tex. The translation was initiated by Algorithms on Mon Jun 2 23:33:50 EDT 1997

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE234.HTM [19/1/2003 1:28:03]

Correctness and Efficiency

Next: Correctness Up: Introduction to Algorithms Previous: Introduction to Algorithms

Correctness and Efficiency Throughout this book we will seek algorithms that are correct and efficient, while being easy to implement. All three goals are obviously desirable, but they may not be simultaneously achievable. For this reason, one or more of them are often ignored. Theoretical algorithm designers have traditionally been unconcerned with implementation complexity, since they often do not program their algorithms. Instead, theoreticians focus on efficiency and correctness. Conversely, quick-and-dirty is typically the rule of thumb in industrial settings. Any program that seems to give good enough answers without slowing the application down is acceptable, regardless of whether a better algorithm exists. The issue of finding the best possible answer or achieving maximum efficiency usually does not arise in industry until serious troubles do. Here, we stress the importance of recognizing the difference between algorithms, which always produce a correct result, and heuristics, which often do a good job without providing any guarantee. We also emphasize the potential efficiency gains resulting from using faster algorithms.

● ●

Correctness Efficiency

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE7.HTM [19/1/2003 1:28:04]

Correctness

Next: Efficiency Up: Correctness and Efficiency Previous: Correctness and Efficiency

Correctness

It is seldom obvious whether a given algorithm correctly solves a given problem. This is why correct algorithms usually come with a proof of correctness, which is an explanation of why we know that the algorithm correctly takes all instances of the problem to the desired result. In this book, we will not stress formal proofs of correctness, primarily because they take substantial mathematical maturity to properly appreciate and construct. However, before we go further it is important to demonstrate why ``it's obvious'' never suffices as a proof of correctness and usually is flat-out wrong. To illustrate, let us consider a problem that often arises in manufacturing, transportation, and testing applications. Suppose we are given a robot arm equipped with a tool, say a soldering iron. In manufacturing circuit boards, all the chips and other components must be fastened onto the substrate. More specifically, each chip has a set of contact points (or wires) that must be soldered to the board. To program the robot arm to perform this soldering job, we must first construct an ordering of the contact points, so that the robot visits (and solders) the first contact point, then visits the second point, third, and so forth until the job is done. The robot arm must then proceed back to the first contact point to prepare for the next board, thus turning the tool-path into a closed tour, or cycle. Since robots are expensive devices, we want to find the tour that minimizes the time it takes to assemble the circuit board. A reasonable assumption is that the robot arm moves with fixed speed, so that the time it takes to travel between two points is the same as its distance. In short, we must solve the following algorithm problem: Input: A set S of n points in the plane. Output: What is the shortest cycle tour that visits each point in the set S? You are given the job of programming the robot arm. Stop right now and think about an algorithm to solve this problem. I'll be happy to wait until you find one.

Several possible algorithms might come to mind to solve this problem. Perhaps the most popular idea is the nearest-neighbor heuristic. Starting from some point , we walk first to its nearest neighbor . From , we walk to its nearest unvisited neighbor, thus excluding only as a candidate. We now repeat this process until we run out of unvisited points, at which time we return to to close off the tour. In pseudocode, the nearest-neighbor heuristic looks like this:

Figure: A good example for the nearest neighbor heuristic

file:///E|/BOOK/BOOK/NODE8.HTM (1 of 5) [19/1/2003 1:28:06]

Correctness

NearestNeighborTSP(P)

Pick and visit an initial point

from P

i = 0

While there are still unvisited points

i = i+1

Select

to be the closest unvisited point to

Visit

Return to

from

This algorithm has a lot to recommend it. It is simple to understand and implement. It makes sense to visit nearby points before we visit faraway points if we want to minimize the total travel time. The algorithm works perfectly on the example in Figure . The nearest neighbor rule is very efficient, for it looks at each pair of points at most twice, once when adding to the tour, the other when adding . Against all these positives there is only one problem. This algorithm is completely wrong.

Figure: A bad example for the nearest neighbor heuristic Wrong? How can it be wrong? The algorithm always finds a tour, but the trouble is that it doesn't necessarily find the shortest possible tour. It doesn't necessarily even come close. Consider the set of points in Figure , all of which lie spaced along a line. The numbers describe the distance that each point lies to the left or right of the point labeled `0'. When we start from the point `0' and repeatedly walk to the nearest unvisited neighbor, you will see that we keep jumping left-right-left-right over `0'. A much better (indeed optimal) tour for these points starts from the leftmost point and visits each point as we walk right before returning at the rightmost point. Try now to imagine your boss's delight as she watches a demo of your robot arm hopscotching left-right-left-right during the assembly of such a simple board. ``But wait,'' you might be saying. ``The problem was in starting at point `0'. Instead, why don't we always start the nearestneighbor rule using the leftmost point as the starting point ? By doing this, we will find the optimal solution on this file:///E|/BOOK/BOOK/NODE8.HTM (2 of 5) [19/1/2003 1:28:06]

Correctness

example.'' That is 100% true, at least until we rotate our example 90 degrees. Now all points are equally leftmost. If the point `0' were moved just slightly to the left, it would be picked as the starting point. Now the robot arm will hopscotch up-down-up-down instead of left-right-left-right, but the travel time will be just as bad as before. No matter what you do to pick the first point, the nearest neighbor rule is doomed not to work correctly on certain point sets. Maybe what we need is a different approach. Always walking to the closest point is too restrictive, since it seems to trap us into making moves we didn't want. A different idea would be to repeatedly connect the closest pair of points whose connection will not cause a cycle or a three-way branch to be formed. Eventually, we will end up with a single chain containing all the points in it. At any moment during the execution of this closest-pair heuristic, we will have a set of partial paths to merge, where each vertex begins as its own partial path. Connecting the final two endpoints gives us a cycle. In pseudocode:

ClosestPairTSP(P)

Let n be the number of points in set P.

For i=1 to n-1 do

For each pair of endpoints (,t) of distinct partial paths

if

then

,

,

and d = dist(,t)

Connect

by an edge

Connect the two endpoints by an edge

This closest-pair rule does the right thing on the example in Figure . It starts by connecting `0' to its immediate neighbors, the points 1 and -1. Subsequently, the next closest pair will alternate left-right, growing the central path by one link at a time. The closest-pair heuristic is somewhat more complicated and less efficient than the previous one, but at least it gives the right answer. On this example.

file:///E|/BOOK/BOOK/NODE8.HTM (3 of 5) [19/1/2003 1:28:06]

Correctness

Figure: A bad example for the closest-pair heuristic. But not all examples. Consider what the algorithm does on the point set in Figure (a). It consists of two rows of equally spaced points, with the rows slightly closer together (distance 1-e) than the neighboring points are spaced within each row (distance 1+e). Thus the closest pairs of points stretch across the gap, not around the boundary. After we pair off these points, the closest remaining pairs will connect these pairs alternately around the boundary. The total path length of the closest-pair . Compared to the tour shown in Figure

tour is than necessary when

(b), we travel over 20% farther

. Examples exist where the penalty is considerably worse than this.

Thus this second algorithm is also wrong. Which one of these algorithms performs better? You can't tell just by looking at them. Clearly, both heuristics can end up with very bad tours on very innocent-looking input. At this point, you might wonder what a correct algorithm for the problem looks like. Well, we could try all enumerating all possible orderings of the set of points, and then select the ordering that minimizes the total length:

OptimalTSP(P)

For each of the n! permutations

If

of points P

then

and

Return

Since all possible orderings are considered, we are guaranteed to end up with the shortest possible tour. This algorithm is correct, since we pick the best of all the possibilities. The problem is that it is also extremely slow. The fastest computer in the world couldn't hope to enumerate all 20! = 2,432,902,008,176,640,000 orderings of 20 points within a day. For real circuit boards, where , forget about it. All of the world's computers working full time wouldn't come close to finishing your problem before the end of the universe, at which point it presumably becomes moot. The quest for an efficient algorithm to solve this so-called traveling salesman problem will take us through much of this book. If you need to know how the story ends, check out the catalog entry for the traveling salesman problem in Section

.

Hopefully, this example has opened your eyes to some of the subtleties of algorithm correctness. Ensuring the optimal answer file:///E|/BOOK/BOOK/NODE8.HTM (4 of 5) [19/1/2003 1:28:06]

Correctness

on all possible inputs is a difficult but often achievable goal. Seeking counterexamples that break pretender algorithms is a tricky business, but it is an important part of the algorithm design process.

Next: Efficiency Up: Correctness and Efficiency Previous: Correctness and Efficiency Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE8.HTM (5 of 5) [19/1/2003 1:28:06]

Efficiency

Next: Expressing Algorithms Up: Correctness and Efficiency Previous: Correctness

Efficiency The skilled algorithm designer is an efficiency expert. Through cunning and experience we seek to get jobs done as quickly as possible. The benefits of finding an efficient algorithm for a computationally expensive job can be mind-boggling, as illustrated by several of the war stories scattered through this book. Section initial attempt!

shows how clever algorithms yielded a program that was 30,000 times faster than our

Upon realizing that a given application runs too slowly, the practical person usually asks the boss to buy them a faster machine. Sometimes this is the cheapest option. However, the potential win from faster hardware is typically limited to a factor of ten or so, due to technical constraints (they don't make faster machines) or economic ones (the boss won't pay for it). To realize larger performance improvements, we must seek better algorithms. As we will show, a faster algorithm running on a slower computer will always win for sufficiently large instances. Always. Usually, problems don't have to get very large before the faster algorithm wins. Be aware that there are situations where finding the most efficient algorithm for a job is a complete waste of programmer effort. In any program, there is usually one bottleneck that takes the majority of computing time. The typical claim is that 90% of the run time of any program is spent in 10% of the code. Optimizing the other 90% of the program will have little impact on the total run time. Further, many programs are written with only one or two special instances in mind. If the program will be run just a few times, or if the job can easily be run overnight, it probably does not pay to make the programmer work harder in order to reduce cycle consumption.

Next: Expressing Algorithms Up: Correctness and Efficiency Previous: Correctness Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE9.HTM [19/1/2003 1:28:06]

Expressing Algorithms

Next: Keeping Score Up: Introduction to Algorithms Previous: Efficiency

Expressing Algorithms Describing algorithms requires a notation for expressing a sequence of steps to be performed. The three most common options are (1) English, (2) pseudocode, or (3) a real programming language. Pseudocode is perhaps the most mysterious of the bunch, but it is best defined as a programming language that never complains about syntax errors. All three methods can be useful in certain circumstances, since there is a natural tradeoff between greater ease of expression and precision. English is the most natural but least precise language, while C and Pascal are precise but difficult to write and understand. Pseudocode is useful because it is a happy medium. The correct choice of which notation is best depends upon which of the three methods you are most comfortable with. I prefer to describe the ideas of an algorithm in English, moving onto a more formal, programming-language-like pseudocode to clarify sufficiently tricky details of the algorithm. A common mistake among my students is to use pseudocode to take an ill-defined idea and dress it up so that it looks more formal. In the real world, you only fool yourself when you pull this kind of stunt. The implementation complexity of an algorithm is usually why the fastest algorithm known for a problem may not be the most appropriate for a given application. An algorithm's implementation complexity is often a function of how it has been described. Fast algorithms often make use of very complicated data structures, or use other complicated algorithms as subroutines. Turning these algorithms into programs requires building implementations of every substructure. Each catalog entry in Section points out available implementations of algorithms to solve the given problem. Hopefully, these can be used as building blocks to reduce the implementation complexity of algorithms that use them, thus making more complicated algorithms worthy of consideration in practice.

Next: Keeping Score Up: Introduction to Algorithms Previous: Efficiency Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE10.HTM [19/1/2003 1:28:07]

Keeping Score

Next: The RAM Model of Up: Introduction to Algorithms Previous: Expressing Algorithms

Keeping Score Algorithms are the most important, durable, and original part of computer science because they can be studied in a language- and machine-independent way. This means that we need techniques that enable us to compare algorithms without implementing them. Our two most important tools are (1) the RAM model of computation and (2) asymptotic analysis of worst-case complexity. This method of keeping score will be the most mathematically demanding part of this book. However, it is important to understand why we need both of these tools to analyze the performance of algorithms. Once you understand the intuition behind these ideas, the formalism becomes a lot easier to deal with.

● ●

The RAM Model of Computation Best, Worst, and Average-Case Complexity

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE11.HTM [19/1/2003 1:28:07]

The RAM Model of Computation

Next: BestWorst, and Average-Case Up: Keeping Score Previous: Keeping Score

The RAM Model of Computation Machine-independent algorithm design depends upon a hypothetical computer called the Random Access Machine or RAM. Under this model of computation, we are confronted with a computer where: ● ●

●

Each ``simple'' operation (+, *, -, =, if, call) takes exactly 1 time step. Loops and subroutines are not considered simple operations. Instead, they are the composition of many single-step operations. It makes no sense for ``sort'' to be a single-step operation, since sorting 1,000,000 items will take much longer than sorting 10 items. The time it takes to run through a loop or execute a subprogram depends upon the number of loop iterations or the specific nature of the subprogram. Each memory access takes exactly one time step, and we have as much memory as we need. The RAM model takes no notice of whether an item is in cache or on the disk, which simplifies the analysis.

Under the RAM model, we measure the run time of an algorithm by counting up the number of steps it takes on a given problem instance. By assuming that our RAM executes a given number of steps per second, the operation count converts easily to the actual run time. The RAM is a simple model of how computers perform. A common complaint is that it is too simple, that these assumptions make the conclusions and analysis too coarse to believe in practice. For example, multiplying two numbers takes more time than adding two numbers on most processors, which violates the first assumption of the model. Memory access times differ greatly depending on whether data sits in cache or on the disk, thus violating the third assumption. Despite these complaints, the RAM is an excellent model for understanding how an algorithm will perform on a real computer. It strikes a fine balance by capturing the essential behavior of computers while being simple to work with. We use the RAM model because it is useful in practice. Every model has a size range over which it is useful. Take, for example, the model that the earth is flat. You might argue that this is a bad model, since the earth is not flat. However, when laying the foundation of a house, the flat earth model is sufficiently accurate that it can be reliably used. Further, it is so much easier to manipulate a flat-earth model that it is inconceivable that you would try to think spherically when you don't have to. The same situation is true with the RAM model of computation. We make an abstraction that in general is very useful. It is quite difficult to design an algorithm such that the RAM model will give you file:///E|/BOOK/BOOK/NODE12.HTM (1 of 2) [19/1/2003 1:28:08]

The RAM Model of Computation

substantially misleading results, by performing either much better or much worse in practice than the model suggests. The robustness of the RAM enables us to analyze algorithms in a machine-independent way.

Next: BestWorst, and Average-Case Up: Keeping Score Previous: Keeping Score Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE12.HTM (2 of 2) [19/1/2003 1:28:08]

Best, Worst, and Average-Case Complexity

Next: The Big Oh Notation Up: Keeping Score Previous: The RAM Model of

Best, Worst, and Average-Case Complexity Using the RAM model of computation, we can count how many steps our algorithm will take on any given input instance by simply executing it on the given input. However, to really understand how good or bad an algorithm is, we must know how it works over all instances. To understand the notions of the best, worst, and average-case complexity, one must think about running an algorithm on all possible instances of data that can be fed to it. For the problem of sorting, the set of possible input instances consists of all the possible arrangements of all the possible numbers of keys. We can represent every input instance as a point on a graph, where the x-axis is the size of the problem (for sorting, the number of items to sort) and the y-axis is the number of steps taken by the algorithm on this instance. Here we assume, quite reasonably, that it doesn't matter what the values of the keys are, just how many of them there are and how they are ordered. It should not take longer to sort 1,000 English names than it does to sort 1,000 French names, for example.

Figure: Best, worst, and average-case complexity As shown in Figure , these points naturally align themselves into columns, because only integers represent possible input sizes. After all, it makes no sense to ask how long it takes to sort 10.57 items. Once we have these points, we can define three different functions over them: ●

The worst-case complexity of the algorithm is the function defined by the maximum number of steps taken on any instance of size n. It represents the curve passing through the highest point of

file:///E|/BOOK/BOOK/NODE13.HTM (1 of 2) [19/1/2003 1:28:09]

Best, Worst, and Average-Case Complexity

●

●

each column. The best-case complexity of the algorithm is the function defined by the minimum number of steps taken on any instance of size n. It represents the curve passing through the lowest point of each column. Finally, the average-case complexity of the algorithm is the function defined by the average number of steps taken on any instance of size n.

In practice, the most useful of these three measures proves to be the worst-case complexity, which many people find counterintuitive. To illustrate why worst-case analysis is important, consider trying to project what will happen to you if you bring n dollars to gamble in a casino. The best case, that you walk out owning the place, is possible but so unlikely that you should place no credence in it. The worst case, that you lose all n dollars, is easy to calculate and distressingly likely to happen. The average case, that the typical person loses 87.32% of the money that they bring to the casino, is difficult to compute and its meaning is subject to debate. What exactly does average mean? Stupid people lose more than smart people, so are you smarter or dumber than the average person, and by how much? People who play craps lose more money than those playing the nickel slots. Card counters at blackjack do better on average than customers who accept three or more free drinks. We avoid all these complexities and obtain a very useful result by just considering the worst case. The important thing to realize is that each of these time complexities defines a numerical function, representing time versus problem size. These functions are as well-defined as any other numerical function, be it or the price of General Motors stock as a function of time. Time complexities are complicated functions, however. In order to simplify our work with such messy functions, we will need the big Oh notation.

Next: The Big Oh Notation Up: Keeping Score Previous: The RAM Model of Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE13.HTM (2 of 2) [19/1/2003 1:28:09]

The Big Oh Notation

Next: Growth Rates Up: Introduction to Algorithms Previous: BestWorst, and Average-Case

The Big Oh Notation We have agreed that the best, worst, and average-case complexities of a given algorithm are numerical functions of the size of the instances. However, it is difficult to work with these functions exactly because they are often very complicated, with many little up and down bumps, as shown in Figure Thus it is usually cleaner and easier to talk about upper and lower bounds of such functions. This is where the big Oh notation comes into the picture.

.

Figure: Upper and lower bounds smooth out the behavior of complex functions We seek this smoothing in order to ignore levels of detail that do not impact our comparison of algorithms. Since running our algorithm on a machine that is twice as fast will cut the running times of all algorithms by a multiplicative constant of two, such constant factors would be washed away in upgrading machines. On the RAM we ignore such constant factors anyway and therefore might as well ignore them when comparing two algorithms. We use the following notation: ●

●

●

f(n) = O(g(n)) means is an upper bound on f(n). Thus there exists some constant c such that f(n) is always , for large enough n. means is a lower bound on f(n). Thus there exists some constant c such that f(n) is always , for large enough n. means is an upper bound on f(n) and is a lower bound on f(n), for large enough n. Thus there exists constants and such that and . This means that g(n) is a nice, tight bound on f(n).

file:///E|/BOOK/BOOK/NODE14.HTM (1 of 2) [19/1/2003 1:28:11]

The Big Oh Notation

Figure: Illustrating the (a) O, (b)

, and (c)

notations

Got it? These definitions are illustrated in Figure . All of these definitions imply a constant beyond which they are always satisfied. We are not concerned about small values of n, i.e. anything to the left of . After all, do you really care whether one sorting algorithm sorts six items faster than another algorithm, or which one is faster when sorting 1,000 or 1,000,000 items? The big Oh notation enables us to ignore details and focus on the big picture. Make sure you understand this notation by working through the following examples. We choose certain constants in the explanations because they work and make a point, but you are free to choose any constant that maintains the same inequality:

The big Oh notation provides for a rough notion of equality when comparing functions. It is somewhat jarring to see an expression like , but its meaning can always be resolved by going back to the definitions in terms of upper and lower bounds. It is perhaps most instructive to read `=' as meaning one . of the functions that are. Clearly, is one of functions that are

Next: Growth Rates Up: Introduction to Algorithms Previous: BestWorst, and Average-Case Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE14.HTM (2 of 2) [19/1/2003 1:28:11]

Growth Rates

Next: Logarithms Up: Introduction to Algorithms Previous: The Big Oh Notation

Growth Rates In working with the big Oh notation, we cavalierly discard the multiplicative constants. The functions and are treated identically, even though g(n) is a million times larger than f(n) for all values of n. The method behind this madness is illustrated by Figure , which tabulates the growth rate of several functions arising in algorithm analysis, on problem instances of reasonable size. shows how long it takes for algorithms that use f(n) operations to complete on a Specifically, Figure fast computer where each operation takes one nanosecond ( seconds). Study the table for a few minutes and the following conclusions become apparent:

Figure: Growth rates of common functions measured in nanoseconds ● ● ●

●

● ●

All of these algorithms take about the same amount of time for n=10. The algorithm whose running time is n! becomes useless well before n=20. The algorithm whose running time is has a greater operating range, but it becomes impractical for n > 40. The algorithm whose running time is is perfectly reasonable up to about n=100, but it quickly deteriorates with larger inputs. For n > 1,000,000 it likely to be hopeless. Both the n and algorithms remain practical on inputs of up to one billion items. You can't hope to find a real problem where an algorithm is going to be too slow in practice.

The bottom line is that even by ignoring constant factors, we can get an excellent idea of whether a given algorithm will be able to run in a reasonable amount of time on a problem of a given size. An algorithm whose running time is seconds will beat one whose running time is seconds . Such enormous constant factor differences between algorithms occur in only so long as practice far less frequently than such large problems do.

file:///E|/BOOK/BOOK/NODE15.HTM (1 of 2) [19/1/2003 1:28:12]

Logarithms

Next: Modeling the Problem Up: Introduction to Algorithms Previous: Growth Rates

Logarithms Logarithm is an anagram of algorithm, but that's not why the wily algorist needs to know what logarithms are and where they come from. You've seen the button on your calculator but have likely is forgotten why it is there. A logarithm is simply an inverse exponential function. Saying . Exponential functions are functions that grow at a distressingly fast equivalent to saying that rate, as anyone who has ever tried to pay off a mortgage or bank loan understands. Thus inverse exponential functions, i.e. logarithms, grow refreshingly slowly. If you have an algorithm that runs in time, take it and run. As shown by Figure problem instances.

, this will be blindingly fast even on very large

Binary search is an example of an algorithm that takes time. In a telephone book with n names, you start by comparing the name that you are looking for with the middle, or (n/2)nd name, say Monroe, Marilyn. Regardless of whether you are looking someone before this middle name (Dean, James) or after it (Presley, Elvis), after only one such comparison you can forever disregard one half of all the names in the book. The number of steps the algorithm takes equals the number of times we can halve n before only one name is left. By definition, this is exactly . Thus twenty comparisons suffice to find any name in the million-name Manhattan phone book! The power of binary search and logarithms is one of the most fundamental ideas in the analysis of algorithms. This power becomes apparent if we could imagine living in a world with only unsorted telephone books. Figure is another example of logarithms in action. This table appears in the Federal Sentencing Guidelines, used by courts throughout the United States. These guidelines are an attempt to standardize criminal sentences, so that a felon convicted of a crime before one judge receives the same sentence as they would before a different judge. To accomplish this, the judges have prepared an intricate point function to score the depravity of each crime and map it to time-to-serve.

file:///E|/BOOK/BOOK/NODE16.HTM (1 of 3) [19/1/2003 1:28:14]

Logarithms

Figure: The Federal Sentencing Guidelines for Fraud Figure gives the actual point function for fraud, a table mapping dollars stolen to points. Notice that the punishment increases by one level each time the amount of money stolen roughly doubles. That means that the level of punishment (which maps roughly linearly to the amount of time served) grows logarithmically with the amount of money stolen. Think for a moment about the consequences of this. Michael Milken sure did. It means that the total sentence grows extremely slowly with the amount of money you steal. Knocking off five liquor stores for $10,000 each will get you far more time than embezzling $100,000 once. The corresponding benefit of stealing really large amounts of money is even greater. The moral of logarithmic growth is clear: ``If you are gonna do the crime, make it worth the time!'' Two mathematical properties of logarithms are important to understand: ●

The base of the logarithm has relatively little impact on the growth rate: Compare the following , , and .A three values: big change in the base of the logarithm produces relatively little difference in the value of the log. This is a consequence of the formula for changing the base of the logarithm:

Changing the base of the log from a to c involves multiplying or dividing by . This will be lost to the big Oh notation whenever a and c are constants, as is typical. Thus we are usually

file:///E|/BOOK/BOOK/NODE16.HTM (2 of 3) [19/1/2003 1:28:14]

Logarithms

justified in ignoring the base of the logarithm when analyzing algorithms.

●

When performing binary search in a telephone book, how important is it that each query split the book exactly in half? Not much. Suppose we did such a sloppy job of picking our queries such that each time we split the book 1/3 to 2/3 instead of 1/2 to 1/2. For the Manhattan telephone book, we now use queries in the worst case, not a significant change from . The power of binary search comes from its logarithmic complexity, not the base of the log. Logarithms cut any function down to size: The growth rate of the logarithm of any polynomial function is . This follows because

The power of binary search on a wide range of problems is a consequence of this observation. For example, note that doing a binary search on a sorted array of things requires only twice as many comparisons as a binary search on n things. Logarithms efficiently cut any function down to size.

Next: Modeling the Problem Up: Introduction to Algorithms Previous: Growth Rates Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE16.HTM (3 of 3) [19/1/2003 1:28:14]

Modeling the Problem

Next: About the War Stories Up: Introduction to Algorithms Previous: Logarithms

Modeling the Problem Modeling is the art of formulating your application in terms of precisely described, well-understood problems. Proper modeling is the key to applying algorithmic design techniques to any real-world problem. Indeed, proper modeling can eliminate the need to design or even implement algorithms by relating your application to what has been done before. Proper modeling is the key to effectively using the problem catalog in Part II of this book. Real-world applications involve real-world objects. You might be working on a system to route traffic in a network, to find the best way to schedule classrooms in a university, or to search for patterns in a corporate database. Most algorithms, however, are designed to work on rigorously defined abstract structures such as permutations, graphs, and sets. After all, if you can't define what you want to do, you can't hope to compute it. You must first describe your problem abstractly, in terms of fundamental structures and properties. Whatever your application is, odds are very good that others before you have stumbled upon the same algorithmic problem, perhaps in substantially different contexts. Therefore, to find out what is known about your particular ``widget optimization problem,'' you can't hope to look in a book under widget. You have to formulate widget optimization in terms of computing properties of abstract structures such as: ●

●

Permutations, which are arrangements, or orderings, of items. For example, and are two distinct permutations of the same set of four integers. Permutations are likely the object in question whenever your problem seeks an ``arrangement,'' ``tour,'' ``ordering,'', or ``sequence.'' Subsets, which represent selections from a set of items. For example, and are two distinct subsets of the first four integers. Order does not matter in subsets the way it does with and would be considered identical. Subsets are likely permutations, so the subsets the object in question whenever your problem seeks a ``cluster,'' ``collection,'' ``committee,'' ``group,'' ``packaging,'' or ``selection.''

file:///E|/BOOK/BOOK/NODE17.HTM (1 of 3) [19/1/2003 1:28:15]

Modeling the Problem

Figure: Modeling real-world structures with trees and graphs ●

●

●

●

●

Trees, which represent hierarchical relationships between items. Figure (a) illustrates a portion of the family tree of the Skiena clan. Trees are likely the object in question whenever your problem seeks a ``hierarchy,'' ``dominance relationship,'' ``ancestor/decendant relationship,'' or ``taxonomy.'' Graphs, which represent relationships between arbitrary pairs of objects. Figure (b) models a network of roads as a graph, where the vertices are cities and the edges are roads connecting pairs of cities. Graphs are likely the object in question whenever you seek a ``network,'' ``circuit,'' ``web,'' or ``relationship.'' Points, which represent locations in some geometric space. For example, the locations of McDonald's restaurants can be described by points on a map/plane. Points are likely the object in question whenever your problems work on ``sites,'' ``positions,'' ``data records,'' or ``locations.'' Polygons, which represent regions in some geometric space. For example, the borders of a country can be described by a polygon on a map/plane. Polygons and polyhedra are likely the object in question whenever you are working on ``shapes,'' ``regions,'' ``configurations,'' or ``boundaries.'' Strings, which represent sequences of characters or patterns. For example, the names of students in a class can be represented by strings. Strings are likely the object in question whenever you are dealing with ``text,'' ``characters,'' ``patterns,'' or ``labels.''

These fundamental structures all have associated problems and properties, which are presented in the catalog of Part II. Familiarity with all of these problems is important, because they provide the language we use to model applications. To become fluent in this vocabulary, browse through the catalog and study the input and output pictures for each problem. Understanding all or most of these problems, even at a cartoon/definition level, will enable you to know where to look later when the problem arises in your application. Examples of successful application modeling will be presented in the war stories spaced throughout this book. However, some words of caution are in order. The act of modeling reduces your application to one of a small number of existing problems and structures. Such a process is inherently constraining, and certain details might not fit easily into the given model. Also, certain problems can be modeled in several different ways, some much better than others. Modeling is only the first step in designing an algorithm for a problem. Be alert for how the details of file:///E|/BOOK/BOOK/NODE17.HTM (2 of 3) [19/1/2003 1:28:15]

Modeling the Problem

your applications differ from a candidate model. But don't be too quick to say that your problem is unique and special. Temporarily ignoring details that don't fit can free the mind to ask whether they were fundamental in the first place.

Next: About the War Stories Up: Introduction to Algorithms Previous: Logarithms Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE17.HTM (3 of 3) [19/1/2003 1:28:15]

About the War Stories

Next: War Story: Psychic Modeling Up: Introduction to Algorithms Previous: Modeling the Problem

About the War Stories The best way to become convinced that careful algorithm design can have a significant impact on performance is to look at case studies. By carefully studying other people's experiences, we learn how they might apply to our work. Jon Bentley's Programming Pearls columns are probably the most interesting collection of algorithmic ``war stories'' currently available. Originally published in the Communications of the ACM, they have been collected in two books [Ben86, Ben90]. Fredrick Brooks's The Mythical Man Month [Bro74] is another wonderful collection of war stories, focused more on software engineering than algorithm design, but they remain a source of considerable wisdom. Every programmer should read all three of these books, for pleasure as well as insight. Scattered throughout this text are several of our own war stories, presenting our successful (and occasionally unsuccessful) algorithm design efforts on real applications. We hope that the reader will be able to internalize these experiences so that they will serve as models for their own attacks on problems. Every one of the war stories is true. Of course, the stories improve somewhat in the retelling, and the dialogue has been punched up to make it more interesting to read. However, I have tried to honestly trace the process of going from a raw problem to a solution, so you can watch how the process unfolded. The Oxford English Dictionary defines an algorist as ``one skillful in reckonings or figuring.'' In these stories, I have tried to capture some of the attitude, or mindset, of the algorist in action as they attack a problem. The various war stories usually involve at least one and often several problems from the problem catalog in Part II. The appropriate section of the catalog is referenced wherever it occurs. This emphasizes the benefits of modeling your application in terms of standard algorithm problems. By using the catalog, you will be able to pull out what is known about any problem whenever it is needed.

file:///E|/BOOK/BOOK/NODE18.HTM (1 of 2) [19/1/2003 1:28:16]

About the War Stories

Next: War Story: Psychic Modeling Up: Introduction to Algorithms Previous: Modeling the Problem Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE18.HTM (2 of 2) [19/1/2003 1:28:16]

War Story: Psychic Modeling

Next: Exercises Up: Introduction to Algorithms Previous: About the War Stories

War Story: Psychic Modeling The call came for me out of the blue as I sat in my office. ``Professor Skiena, I hope you can help me. I'm the President of Lotto Systems Group Inc., and we need an algorithm for a problem arising in our latest product.'' ``Sure,'' I replied. After all, the dean of my engineering school is always eager for our faculty to interact with industry. ``At Lotto Systems Group, we market a program designed to improve our customers' psychic ability to predict winning lottery numbers. In a standard lottery, each ticket consists of 6 numbers selected from, say, 1 to 44. Thus any given ticket has only a very small chance of winning. However, after proper training, our clients can visualize a set of, say, 15 numbers out of the 44 and be certain that at least 4 of them will be on the winning ticket. Are you with me so far?'' ``Probably not,'' I replied. But then I recalled how my dean wants us to interact with industry. ``Our problem is this. After the psychic has narrowed the choices down to 15 numbers and is certain that at least 4 of them will be on the winning ticket, we must find the most efficient way to exploit this information. Suppose that a cash prize is awarded whenever you pick at least three of the correct numbers on your ticket. We need an algorithm to construct the smallest set of tickets that we must buy in order to guarantee that we win at least one prize.'' ``Assuming the psychic is correct?'' ``Yes, assuming the psychic is correct. We need a program that prints out a list of all the tickets that the psychic should buy in order to minimize their investment. Can you help us?'' I thought about the problem for a minute. Maybe they did have psychic ability, for they had clearly come to the right place. Identifying the best subset of tickets to buy was very much a combinatorial algorithm problem. It was going to be some type of covering problem, where each ticket we buy was going to ``cover'' some of the possible 4-element subsets of the psychic's set. Finding the absolute smallest set of tickets to cover everything was a special instance of the NP-complete problem set cover (discussed in Section

) and presumably computationally intractable.

file:///E|/BOOK/BOOK/NODE19.HTM (1 of 5) [19/1/2003 1:28:18]

War Story: Psychic Modeling

Figure: Covering all pairs of

with tickets

,

,

,

It was indeed a special instance of set cover, completely specified by only four numbers: the size n of the candidate set S (typically ), the number of k slots for numbers on each ticket (typically ), the number of psychically promised correct numbers j from S (say j=4), and finally, the number of numbers l matching the winning ticket necessary to win a prize (say l=3). Figure

illustrates a covering of a smaller instance, where n=5, k=3, and j=l=2.

``Although it will be hard to find the exact minimum set of tickets to buy, with heuristics I should be able to get you pretty close to the cheapest covering ticket set,'' I told him. ``Will that be good enough?'' ``So long as it generates better ticket sets than my competitor's program, that will be fine. His system doesn't always provide enough coverage to guarantee a win. I really appreciate your help on this, Professor Skiena.'' ``One last thing. If your program can train people to pick lottery winners, why don't you use it to win the lottery yourself?'' ``I look forward to talking to you again real soon, Professor Skiena. Thanks for the help.'' I hung up the phone and got back to thinking. It seemed like a fun problem, and the perfect project to give to a bright undergraduate. After it was modeled in terms of sets and subsets, the basic components of a solution seemed fairly straightforward: ●

We needed the ability to generate all subsets of k numbers from the candidate set S. Algorithms for generating and

●

. ranking/unranking subsets of sets are presented in Section We needed the right formulation of what it meant to have a covering set of purchased tickets. The obvious criteria would be to pick a small set of tickets such that for each of the l-subsets of S that might pay off with the prize, we

file:///E|/BOOK/BOOK/NODE19.HTM (2 of 5) [19/1/2003 1:28:18]

War Story: Psychic Modeling

●

●

have purchased at least one ticket that contains it. For each new ticket we decided to buy, we would need to keep track of which prize combinations we have thus far covered. Presumably, we would want to buy tickets to cover as many thus-far-uncovered prize combinations as possible. The currently covered combinations are a subset of all possible combinations, so we would need a data . The best candidate seemed to be a structure to represent the subset. Such data structures are discussed in Section bit vector, which would permit a constant time query of ``is this combination already covered?'' in concert with a way to rank the subset, i.e. hash it to a unique integer. We would need a search mechanism to decide which ticket to buy next. For small enough set sizes, we could do an exhaustive search over all possible subsets of tickets and pick the smallest one. For larger sizes, a randomized search ) would select tickets-to-buy to cover as many uncovered process like simulated annealing (see Section combinations as possible. By repeating this randomized procedure several times and picking the best solution, we would be likely to come up with a very good set of tickets.

Excluding the search mechanism, the pseudocode for the bookkeeping looked something like this:

LottoTicketSet(n,k,l)

Initialize the

-element bit-vector V to all false

While there exists a false entry of V

Select a k-subset T of

as the next

ticket to buy

For each of the l-subsets

of T,

Report the set of tickets bought

The bright undergraduate, Fayyaz Younas, rose to the challenge. Based on this framework, he implemented a brute-force in a reasonable time. To solve larger problems, he search algorithm and found optimal solutions for problems with implemented a random search procedure, tweaking it for a while before settling on the best variant. Finally, the day arrived when we could call Lotto Systems Group and announce that we had solved the problem. ``See, our program found that optimal solution for n=15, k=6, j=6, l=3 meant buying 28 tickets.'' ``Twenty-eight tickets!'' complained the president. ``You must have a bug. Look, these five tickets will suffice to cover everything twice over: , , , , .''

file:///E|/BOOK/BOOK/NODE19.HTM (3 of 5) [19/1/2003 1:28:18]

War Story: Psychic Modeling

Figure: Guaranteeing a winning ticket from

using only tickets

and

We fiddled with this example for a while before admitting that he was right. We hadn't modeled the problem correctly! In illustrates the principle by giving a fact, we didn't need to explicitly cover all possible winning combinations. Figure better solution to the previous small example. As a different example, suppose we needed three numbers on a five element . Suppose that we had bought ticket to win, and the psychic had promised us that four numbers would be from , , and . There would be no need to buy a ticket that explicitly tickets that covered the subsets , because if 1 and 2 and two other psychic numbers were on a winning ticket, then at least one of the other covered two psychic numbers would have to be either 3, 4, or 5. If 1 and 2 were not on the winning ticket, a ticket covering would do us absolutely no good. Thus we were trying to cover too many combinations, and the penny-pinching psychics were unwilling to pay for such extravagance. Fortunately, this story has a happy ending, as reported in [YS96]. The general outline of our search-based solution described above still holds for the real problem. All that is needed to fix it is to change how many uncovered subsets we credit for covering with a given set of tickets. After this modification, we obtained the kind of results they were hoping for. Lotto Systems Group gratefully accepted our program to incorporate into their products, and we hope they hit the jackpot with it. The moral of this story is to make sure that you correctly model the problem before trying to solve it. In our case, we came up with a reasonable model for the problem but didn't work hard enough to validate it before we started to program. Our misinterpretation would have become obvious had we worked out a small but non-trivial example by hand and bounced it off our sponsor before beginning work. Our success in recovering from the error is a tribute to the basic correctness of our initial formulation and our use of well-defined routines and abstractions for such tasks as (1) ranking/unranking k-subsets, (2) the set data structure, and (3) combinatorial search.

file:///E|/BOOK/BOOK/NODE19.HTM (4 of 5) [19/1/2003 1:28:18]

War Story: Psychic Modeling

Next: Exercises Up: Introduction to Algorithms Previous: About the War Stories Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE19.HTM (5 of 5) [19/1/2003 1:28:18]

Exercises

Next: Implementation Challenges Up: Introduction to Algorithms Previous: War Story: Psychic Modeling

Exercises 1. Let P be a problem. The worst-case time complexity of P is . The worst-case time . Let A be an algorithm that solves P. Which subset of the complexity of P is also following statements are consistent with this information about the complexity of P? A has worst-case time complexity . ❍ A has worst-case time complexity . ❍ A has worst-case time complexity O(n). ❍ A has worst-case time complexity . ❍ A has worst-case time complexity . 2. Suppose that an algorithm A runs in worst-case time f(n) and that algorithm B runs in worst-case time g(n). For each of the following questions, answer either yes, no, or can't tell and explain why. ❍

(a) Is A faster than B for all n greater than some

if

?

(b) Is A faster than B for all n greater than some

if

?

(c) Is A faster than B for all n greater than some

if

?

(d) Is B faster than A for all n greater than some

if

?

(e) Is B faster than A for all n greater than some

if

?

(f) Is B faster than A for all n greater than some if 3. For each of these questions, briefly explain your answer. (a) If I prove that an algorithm takes some inputs? file:///E|/BOOK/BOOK/NODE20.HTM (1 of 3) [19/1/2003 1:28:20]

?

worst-case time, is it possible that it takes O(n) on

Exercises

(b) If I prove that an algorithm takes inputs?

worst-case time, is it possible that it takes O(n) on all

(c) If I prove that an algorithm takes some inputs?

worst-case time, is it possible that it takes O(n) on

(d) If I prove that an algorithm takes inputs?

worst-case time, is it possible that it takes O(n) on all

(e) Is the function , where for even n and n? 4. For each of the following, answer yes, no, or can't tell. Explain your reasoning! (a) Is

? ?

(b) Is (c) Is

5. 6. 7. 8.

9.

for odd

?

? (d) Is (*) Give a proof or counterexample to the following claim: if f(n) = O(F(n)) and g(n) = O(G(n)), then f(n)/g(n) = O(F(n)/G(n)). (*) Does f(n) = O(g(n)) imply that ? Explain your reasoning! (*) Give a proof or counterexample to the following claim: for all functions f(n) and g(n), either f(n) = O(g(n)) or g(n) = O(f(n)). (*) When you first learned to multiply numbers, you were told that means add x a total of y times, so . What is the time complexity of multiplying two n-digit numbers in base b (people work in base 10, of course, while computers work in base 2) using the repeated addition method, as a function of n and b. Assume that single-digit by single-digit addition or multiplication takes O(1) time. (hint: how big can y be as a function of n and b?) (*) In grade school, you learned to multiply long numbers on a digit by digit basis, so that . Analyze the time complexity of multiplying two n-digit numbers with this method as a function of n (assume constant base size). Assume that single-digit by single-digit addition or multiplication takes O(1) time.

file:///E|/BOOK/BOOK/NODE20.HTM (2 of 3) [19/1/2003 1:28:20]

Exercises

Next: Implementation Challenges Up: Introduction to Algorithms Previous: War Story: Psychic Modeling Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE20.HTM (3 of 3) [19/1/2003 1:28:20]

Fundamental Data Types

Next: Containers Up: Data Structures and Sorting Previous: Data Structures and Sorting

Fundamental Data Types An abstract data type is a collection of well-defined operations that can be performed on a particular structure. The operations define what the data type does, but not how it works. Abstract data types are black boxes that we dare not open when we design algorithms that use them. For each of the most important abstract data types, several competing implementations, or data structures, are available. Often, alternative data structures realize different design tradeoffs that make certain operations (say, insertion) faster at the cost of other operations (say, search). In some applications, certain data structures yield simpler code than other implementations of a given abstract data type, or have some other specialized advantages. We assume that the reader has had some previous exposure to elementary data structures and some fluency in pointer manipulation. Therefore, we do not discuss these topics here. The reader who wants to review elementary data structures is referred to any of the books in Section . Instead, we focus on three fundamental abstract data types: containers, dictionaries, and priority queues. Detailed discussion of the tradeoffs between implementations of these data types is deferred to the relevant catalog entry.

● ● ● ●

Containers Dictionaries Binary Search Trees Priority Queues

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE23.HTM [19/1/2003 1:28:21]

Containers

Next: Dictionaries Up: Fundamental Data Types Previous: Fundamental Data Types

Containers Containers are abstract data types that hold stuff. They don't do much more than hold it so that it can be retrieved later. Still, they are critical to the functioning of society. We will use the term container to denote a data structure that permits storage and retrieval of data items independently of content. The two fundamental operations of any container are: ● ●

Put(C,x): Insert a new data item x into the container C. Get(C): Retrieve the next item from the container C. Different types of containers support different retrieval orders, based on insertion order or position.

Containers are typically most useful when they will contain only a limited number of items and when the retrieval order is predefined or irrelevant. The most popular type of containers are: ●

●

●

Stacks: Supports retrieval in last in, first out order (LIFO). Stacks are simple to implement, and very efficient. Indeed, stacks are probably the right container to use when the retrieval order doesn't matter at all, as when processing batch jobs. The put and get operations for stacks are usually called push and pop. Queues: Supports retrieval in first in, first out order (FIFO). FIFO may seem the fairest way to control waiting times. However, for many applications, data items have infinite patience. Queues are trickier to implement than stacks and are appropriate only for applications (like certain simulations) where the order is important. The put and get operations for queues are usually called enqueue and dequeue. Tables: Supports retrieval by position, so that put and get each accept an index as an argument. Tables are naturally implemented using arrays.

Each of these containers can be implemented using either arrays or linked lists. With the exception of tables, the choice of lists versus tables probably doesn't matter very much. The key issue is whether an upper bound on the size of the container is known in advance, thus permitting a statically allocated array. Using arrays, put and get can be implemented in constant time per operation for each of the containers.

file:///E|/BOOK/BOOK/NODE24.HTM (1 of 2) [19/1/2003 1:28:22]

Containers

Next: Dictionaries Up: Fundamental Data Types Previous: Fundamental Data Types Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE24.HTM (2 of 2) [19/1/2003 1:28:22]

Dictionaries

Next: Binary Search Trees Up: Fundamental Data Types Previous: Containers

Dictionaries

Dictionaries are a form of container that permits access to data items by content. You put a word into a dictionary because you know you can look it up when you need it. The primary operations dictionaries support are: ●

● ●

Search(D,k): Given a search key k, return a pointer to the element in dictionary D whose key value is k, if one exists. Insert(D,x): Given a data item x, add it to the set of items in the dictionary D. Delete(D,x): Given a pointer to a given data item x in the dictionary D, remove it from D.

Perhaps the simplest possible dictionary implementation maintains an unsorted linked list as a data structure. Insertion and deletion are supported in constant time, although a query requires potentially traversing the entire linked list. Basing an implementation on a stored array speeds up the query operation to by binary search. Making room for a new item or filling a hole left by a deletion may require moving arbitrarily many items, so insertion and deletion become linear-time operations. Many other dictionary implementations are available. Binary search trees are discussed in some detail in the next section. Hash tables are another attractive option in practice. A complete discussion of different dictionary data structures is presented catalog Section . We encourage the reader to browse through the data structures section of the catalog in order to learn what your options are. Certain dictionary data structures also efficiently support the following useful operations: ●

●

Max(D) or Min(D): Retrieve the item with the largest (or smallest) key from D. This enables the dictionary to serve as a priority, as discussed below. Predecessor(D,k) or Successor(D,k): Retrieve the item from D whose key is immediately before (or after) k in sorted order. By starting from the first item Min(D) and repeatedly calling Successor until we obtain Max(D), we traverse all elements in sorted order.

file:///E|/BOOK/BOOK/NODE25.HTM (1 of 2) [19/1/2003 1:28:23]

Dictionaries

Next: Binary Search Trees Up: Fundamental Data Types Previous: Containers Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE25.HTM (2 of 2) [19/1/2003 1:28:23]

Binary Search Trees

Next: Priority Queues Up: Fundamental Data Types Previous: Dictionaries

Binary Search Trees Fast support of all dictionary operations is realized by binary search trees. A binary tree is a rooted tree where each node contains at most two children. Each child can be identified as either a left or right child. , a binary tree can be implemented where each node has left and right pointer As shown in Figure fields, an (optional) parent pointer, and a data field.

Figure: Relationships in a binary search tree A binary search tree labels each node in a binary tree with a single key such that for any node labeled x, all nodes in the left subtree of x have keys < x while all nodes in the right subtree of x have keys > x. The search tree labeling enables us to find where any key is. Start at the root. If it does not contain the key we are searching for, proceed either left or right depending upon whether what we want occurs before or after the root key. This algorithm works because both the left and right subtrees of a binary search tree are binary search trees; the recursive structure yields a recursive algorithm. Accordingly, the dictionary Query operation can be performed in O(h) time, where h is the height of the tree.

file:///E|/BOOK/BOOK/NODE26.HTM (1 of 3) [19/1/2003 1:28:24]

Binary Search Trees

BinaryTreeQuery(x, k)

if

then return x

if (k < key[x])

then return BinaryTreeQuery(left[x],k)

else return BinaryTreeQuery(right[x],k)

To insert a new item x with key k into a binary search tree T, it is important to place it where it can be later be found. There is only one such location in any binary search tree, namely by replacing the nil pointer found in T after an unsuccessful query for k. Replacing this nil pointer with a pointer to x is a simple, constant-time operation after the search has been performed in O(h) time. Deletion is somewhat more tricky than insertion, because the node selected to die may not be a leaf. Leaf nodes may be deleted without mercy, by clearing the pointer to the given node. However, internal nodes have children that must remain accessible after the deletion. By restructuring or relabeling the tree, however, the item to delete can always be made into a leaf and then removed. Details appear in any data structures text. When implemented using binary search trees, all three dictionary operations take O(h) time, where h is the height of the tree. The smallest height we could hope for occurs when the tree is perfectly balanced, where . In fact, if we insert the keys in random order, with high probability the tree will have height. However, if we get unlucky with our order of insertion or deletion, we can end up with a linear-height tree in the worst case. The worst case is a serious potential problem. Indeed, it occurs whenever the keys are inserted in sorted order. To avoid such worst-case performance, more sophisticated balanced binary search tree data structures have been developed that guarantee the height of the tree always to be . Therefore, all dictionary operations (insert, delete, query) take time each. Implementations of such balanced tree data structures as red-black trees are discussed in Section file:///E|/BOOK/BOOK/NODE26.HTM (2 of 3) [19/1/2003 1:28:24]

.

Binary Search Trees

From an algorithm design viewpoint, it is most important to know that these trees exist and that they can be used as black boxes to provide an efficient dictionary implementation. When figuring the costs of dictionary operations for algorithm analysis, assume the worst-case complexities of balanced binary trees to be a fair measure.

Next: Priority Queues Up: Fundamental Data Types Previous: Dictionaries Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE26.HTM (3 of 3) [19/1/2003 1:28:24]

Priority Queues

Next: Specialized Data Structures Up: Fundamental Data Types Previous: Binary Search Trees

Priority Queues Many algorithms process items according to a particular order. For example, suppose you have to schedule a list of jobs given the deadline by which each job must be performed or else its importance relative to the other jobs. Scheduling jobs requires sorting them by time or importance, and then performing them in this sorted order. Priority queues provide extra flexibility over sorting, which is required because jobs often enter the system at arbitrary intervals. It is much more cost-effective to insert a new job into a priority queue than to re-sort everything. Also, the need to perform certain jobs may vanish before they are executed, meaning that they must be removed from the queue. The basic priority queue supports three primary operations: ● ●

●

Insert(Q,x): Given an item x with key k, insert it into the priority queue Q. Find-Minimum(Q) or Find-Maximum(Q): Return a pointer to the item whose key value is smaller (larger) than any other key in the priority queue Q. Delete-Minimum(Q) or Delete-Maximum(Q) - Remove the item from the priority queue Q whose key is minimum (maximum).

Figure: The maximum and minimum element in a binary search tree All three of these priority queue operations can be implemented in time by representing the heap with a binary search tree. Implementing the find-minimum operation requires knowing where the minimum element in the tree is. By definition, the smallest key must reside in the left subtree of the root, since all keys in the left subtree have values less than that of the root. Therefore, as shown in Figure , the minimum element must be the leftmost decendent of the root. Similarly, the maximum element must be the rightmost decendent of the root.

Find-Maximum(x) Find-Minimum(x)

while while

do x = right[x] do x = left[x]

return x return x file:///E|/BOOK/BOOK/NODE27.HTM (1 of 2) [19/1/2003 1:28:25]

Priority Queues

Repeatedly traversing left (or right) pointers until we hit a leaf takes time proportional to the height of the tree, or if the tree is balanced. The insert operation can be implemented exactly as binary tree insertion. Delete-Min can be implemented by finding the minimum element and then using standard binary tree deletion. It follows that each of the operations can be performed in time. Priority queues are very useful data structures. Indeed, they are the hero of the war story described in Section set of priority queue implementations is presented in Section

.

Next: Specialized Data Structures Up: Fundamental Data Types Previous: Binary Search Trees Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE27.HTM (2 of 2) [19/1/2003 1:28:25]

. A complete

Specialized Data Structures

Next: Sorting Up: Data Structures and Sorting Previous: Priority Queues

Specialized Data Structures The basic data structures thus far described all represent an unstructured set of items so as to facilitate retrieval operations. These data structures are well known to most programmers. Not as well known are high-powered data structures for representing more structured or specialized kinds of objects, such as points in space, strings, and graphs. The design principles of these data structures are the same as for basic objects. There exists a set of basic operations we need to perform repeatedly. We seek a data structure that supports these operations very efficiently. These efficient, specialized data structures are as important for efficient graph and geometric algorithms as lists and arrays are for basic algorithms, so one should be aware of their existence. Details appear throughout the catalog. ●

●

●

String data structures - Character strings are typically represented by arrays of characters, with perhaps a special character to mark the end of the string. Suffix trees/arrays are special data for structures that preprocess strings to make pattern matching operations faster. See Section details. Geometric data structures - Geometric data typically consists of collections of data points and regions. Regions in the plane are usually described by polygons, where the boundary of the polygon is given by a chain of line segments. Polygons can be represented using an array of points , such that is a segment of the boundary. Spatial data structures such as kd-trees organize points and regions by geometric location to support fast search. See Section for details. Graph data structures - Graphs are typically represented by either adjacency matrices or adjacency lists. The choice of representation can have a substantial impact on the design of the resulting graph algorithms, as discussed in Chapter

●

. Implementation aspects of graph data

. structures are presented in Section Set data structures - Subsets of items are typically represented using a dictionary, to support fast membership queries. A variety of data structures for manipulating sets is presented in Section

file:///E|/BOOK/BOOK/NODE28.HTM (1 of 2) [19/1/2003 1:28:26]

.

Specialized Data Structures

Next: Sorting Up: Data Structures and Sorting Previous: Priority Queues Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE28.HTM (2 of 2) [19/1/2003 1:28:26]

Sorting

Next: Applications of Sorting Up: Data Structures and Sorting Previous: Specialized Data Structures

Sorting By the time they graduate, computer science students are likely to have studied the basic sorting algorithms in their introductory programming class, then in their data structures class, and finally in their algorithms class. Why is sorting worth so much attention? There are several reasons: ●

●

●

●

Sorting is the basic building block around which many other algorithms are built. By understanding sorting, we obtain an amazing amount of power to solve other problems. Historically, computers have spent more time sorting than doing anything else. A quarter of all mainframe cycles are spent sorting data [Knu73b]. Although it is unclear whether this remains true on smaller computers, sorting remains the most ubiquitous combinatorial algorithm problem in practice. Sorting is the most throughly studied problem in computer science. Literally dozens of different algorithms are known, most of which possess some advantage over all other algorithms in certain situations. To become convinced of this, the reader is encouraged to browse through [Knu73b], with hundreds of pages of interesting sorting algorithms and analysis. Most of the interesting ideas used in the design of algorithms appear in the context of sorting, such as divide-and-conquer, data structures, and randomized algorithms.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE29.HTM [19/1/2003 1:28:26]

Applications of Sorting

Next: Approaches to Sorting Up: Data Structures and Sorting Previous: Sorting

Applications of Sorting An important key to algorithm design is to use sorting as a basic building block, because once a set of items is sorted, many other problems become easy. Consider the following applications: ●

●

●

●

●

Searching - Binary search enables you to test whether an item is in a dictionary in time, once the keys are all sorted. Search preprocessing is perhaps the single most important application of sorting. Closest pair - Given a set of n numbers, how do you find the pair of numbers that have the smallest difference between them? After the numbers are sorted, the closest pair of numbers will lie next to each other somewhere in sorted order. Thus a linear-time scan through them completes the job, for a total of time including the sorting. Element uniqueness - Are there any duplicates in a given a set of n items? The most efficient algorithm is to sort them and then do a linear scan though them checking all adjacent pairs. This is a special case of the closest-pair problem above, where we ask if there is a pair separated by a gap of zero. Frequency distribution - Given a set of n items, which element occurs the largest number of times in the set? If the items are sorted, we can sweep from left to right and count them, since all identical items will be lumped together during sorting. To find out how often an arbitrary element k occurs, start by looking up k using binary search in a sorted array of keys. By walking to the left of this point until the element is not k and then walking to the right, we can find this count in time, where c is the number of occurrences of k. The number of instances of k can be found in time by using binary search to look for the positions of both and , where is arbitrarily small, and then taking the difference of these positions. Selection - What is the kth largest item in the set? If the keys are placed in sorted order in an array, the kth largest can be found in constant time by simply looking at the kth position of the array. In particular, the median element (see Section order.

file:///E|/BOOK/BOOK/NODE30.HTM (1 of 2) [19/1/2003 1:28:28]

) appears in the (n/2)nd position in sorted

Applications of Sorting

Figure: Constructing the convex hull of points in the plane ●

Convex hulls - Given n points in two dimensions, what is the polygon of smallest area that contains them all? The convex hull is like a rubber band stretched over the points in the plane and . The convex hull then released. It compresses to just cover the points, as shown in Figure gives a nice representation of the shape of the points and is the most important building block for more sophisticated geometric algorithms, as discussed in catalog Section

.

But how can we use sorting to construct the convex hull? Once you have the points sorted by xcoordinate, the points can be inserted from left to right into the hull. Since the rightmost point is always on the boundary, we know that it will be inserted into the hull. Adding this new rightmost point might cause others to be deleted, but we can quickly identify these points because they lie inside the polygon formed by adding the new point. These points to delete will be neighbors of the previous point we inserted, so they will be easy to find. The total time is linear after the sorting has been done. While some of these problems (particularly median and selection) can be solved in linear time using more sophisticated algorithms, sorting provides quick and easy solutions to all of these problems. It is a rare application whose time complexity is such that sorting proves to be the bottleneck, especially a bottleneck that could have otherwise been removed using more clever algorithmics. Don't ever be afraid to spend time sorting whenever you use an efficient sorting routine.

Next: Approaches to Sorting Up: Data Structures and Sorting Previous: Sorting Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE30.HTM (2 of 2) [19/1/2003 1:28:28]

Approaches to Sorting

Next: Data Structures Up: Data Structures and Sorting Previous: Applications of Sorting

Approaches to Sorting Sorting is a natural laboratory for studying basic algorithm design paradigms, since many useful techniques lead to interesting sorting algorithms. Indeed, we introduce several techniques here that will be described further in subsequent chapters. Consider the algorithms below as case studies for each of the relevant techniques.

● ● ● ● ●

Data Structures Incremental Insertion Divide and Conquer Randomization Bucketing Techniques

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE31.HTM [19/1/2003 1:28:28]

Data Structures

Next: Incremental Insertion Up: Approaches to Sorting Previous: Approaches to Sorting

Data Structures Perhaps the most dramatic algorithmic improvement made possible by using appropriate data structures occurs in sorting. Selection sort is a simple-to-code algorithm that repeatedly extracts the smallest remaining element from the unsorted part of the set:

SelectionSort(A)

For i = 1 to n do

Sort[i] = Find-Minimum from A

Delete-Minimum from A

Return(Sort)

Selection sort is typically implemented by partitioning the input array into sorted and unsorted regions. To find the smallest item, we perform a linear sweep through the unsorted portion of the array. The smallest item is then swapped with the ith item in the array before moving on the next iteration. Selection sort performs n iterations, where the average iteration takes n/2 steps, for a total of time. But what if we improve the data structure? It takes O(1) time to remove a particular item from an unsorted array once it has been located, but O(n) time to find the smallest item. These two are exactly the operations supported by priority queues. So what happens if we replace the data structure with a better priority queue implementation, either a heap or a balanced binary tree. Operations within the loop now take time each, instead of O(n). By using such a priority queue implementation, selection sort is sped up to from . The name typically given to this algorithm, heapsort, obscures the relationship between them, but heapsort is nothing but an implementation of selection sort with the right data structure. file:///E|/BOOK/BOOK/NODE32.HTM (1 of 2) [19/1/2003 1:28:29]

Data Structures

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE32.HTM (2 of 2) [19/1/2003 1:28:29]

Incremental Insertion

Next: Divide and Conquer Up: Approaches to Sorting Previous: Data Structures

Incremental Insertion Now consider a different approach to sorting that grows the sorted set one element at a time. Select an arbitrary element from the unsorted set, and put it in the proper position in the sorted set.

InsertionSort(A)

for i = 1 to n-1 do

j=i

while (A[j] > A[j-1]) do swap(A[j],A[j-1])

Although insertion sort takes in the worst case, it will perform considerably better if the data is almost sorted, since few iterations of the inner loop will suffice to sift it into the proper position. Insertion sort is perhaps the simplest example of the incremental insertion technique, where we build up a complicated structure on n items by first building it on n-1 items and then making the necessary changes to fix things in adding the last item. Incremental insertion proves a particularly useful technique in geometric algorithms.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE33.HTM [19/1/2003 1:28:29]

Divide and Conquer

Next: Randomization Up: Approaches to Sorting Previous: Incremental Insertion

Divide and Conquer

Suppose we take the n elements to sort and split them into piles S and T, each with half the elements. After sorting both piles, it is easy to combine the two sorted piles. To merge and , note that the smallest item must sit at the top of one of the two lists. Once identified, the smallest element can be removed, and the second smallest item will again be atop one of the two lists. Repeating this operation merges the two sorted lists in O(n) time. This provides the basis for a recursive algorithm.

Mergesort(A[1,n])

Merge( MergeSort(

), MergeSort(

) )

levels deep and a linear amount of work is done per level, Mergesort Because the recursion goes time in the worst case. takes Mergesort is a classic divide-and-conquer algorithm. Whenever we can break one large problem into two smaller problems, we are ahead of the game because the smaller problems are easier. The trick is taking advantage of the two partial solutions to put together a solution of the full problem. The merge operation takes care of the reassembly in mergesort, but not all problems can be so neatly decomposed.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE34.HTM [19/1/2003 1:28:30]

Randomization

Next: Bucketing Techniques Up: Approaches to Sorting Previous: Divide and Conquer

Randomization Suppose we select, at random, one of the n items we seek to sort. Call this element p. In quicksort, we separate the n-1 other items into two piles: a low pile containing all the elements that will appear before p in sorted order and a high pile containing all the elements that will appear after p in sorted order. After sorting the two piles and arranging the piles correctly, we have succeeded in sorting the n items.

Quicksort(A, low, high)

if (low < high)

ploc = Partition(A,low,high)

Quicksort(A,low, ploc - 1)

Quicksort(A, ploc+1, high)

Partition(A,low,high)

swap(A[low],A[random(

pivot = A[low]

leftwall = low

file:///E|/BOOK/BOOK/NODE35.HTM (1 of 3) [19/1/2003 1:28:31]

)])

Randomization

for i = low+1 to high

if (A[i] < pivot) then

leftwall = leftwall+1

swap(A[i],A[leftwall])

swap(A[low],A[leftwall])

Mergesort ran in time because we split the keys into two equal halves, sorted them recursively, and then merged the halves in linear time. Thus whenever our pivot element is near the center of the sorted array (i.e. the pivot is close to the median element), we get a good split and realize the same performance as mergesort. Such good pivot elements will often be the case. After all, half the elements lie closer to the middle than one of the ends. On average, Quicksort runs in time. If we are extremely unlucky and our randomly selected elements always are among the largest or smallest element in the array, Quicksort turn into selection sort and runs in . However, the odds against this are vanishingly small. Randomization is a powerful, general tool to improve algorithms with bad worst-case but good averagecase complexity. The worst case examples still exist, but they depend only upon how unlucky we are, not on the order that the input data is presented to us. For randomly chosen pivots, we can say that ``Randomized quicksort runs in

time on any input, with high probability.''

If instead, we used some deterministic rule for selecting pivots (like always selecting A[(low+high)/2] as pivot), we can make no claim stronger than ``Quicksort runs in

time if you give me random input data, with high probability.''

Randomization can also be used to drive search techniques such as simulated annealing, which are discussed in Section

.

file:///E|/BOOK/BOOK/NODE35.HTM (2 of 3) [19/1/2003 1:28:31]

Randomization

Next: Bucketing Techniques Up: Approaches to Sorting Previous: Divide and Conquer Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE35.HTM (3 of 3) [19/1/2003 1:28:31]

Bucketing Techniques

Next: War Story: Stripping Triangulations Up: Approaches to Sorting Previous: Randomization

Bucketing Techniques If we were sorting names for the telephone book, we could start by partitioning the names according to the first letter of the last name. That will create 26 different piles, or buckets, of names. Observe that any name in the J pile must occur after every name in the I pile but before any name in the K pile. Therefore, we can proceed to sort each pile individually and just concatenate the bunch of piles together. If the names are distributed fairly evenly among the buckets, as we might expect, the resulting 26 sorting problems should each be substantially smaller than the original problem. Further, by now partitioning each pile based on the second letter of each name, we generate smaller and smaller piles. The names will be sorted as soon as each bucket contains only a single name. The resulting algorithm is commonly called bucketsort or distribution sort. Bucketing is a very effective idea whenever we are confident that the distribution of data will be roughly uniform. It is the idea that underlies hash tables, kd-trees, and a variety of other practical data structures. The downside of such techniques is that the performance can be terrible whenever the data distribution is not what we expected. Although data structures such as binary trees offer guaranteed worst-case behavior for any input distribution, no such promise exists for heuristic data structures on unexpected input distributions.

Figure: A small subset of Charlottesville Shiffletts To show that non-uniform distributions occur in real life, consider Americans with the uncommon last name of Shifflett. The 1997 Manhattan telephone directory, with over one million names, contains exactly five Shiffletts. So how many Shiffletts shows a small portion of the two and a half pages of Shiffletts in should there be in a small city of 50,000 people? Figure the Charlottesville, Virginia telephone book. The Shifflett clan is a fixture of the region, but it would play havoc with any distribution sort program, as refining buckets from S to Sh to Shi to Shif to to Shifflett results in no significant partitioning.

Next: War Story: Stripping Triangulations Up: Approaches to Sorting Previous: Randomization

file:///E|/BOOK/BOOK/NODE36.HTM (1 of 2) [19/1/2003 1:28:32]

Bucketing Techniques

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE36.HTM (2 of 2) [19/1/2003 1:28:32]

War Story: Stripping Triangulations

Next: War Story: Mystery of Up: Data Structures and Sorting Previous: Bucketing Techniques

War Story: Stripping Triangulations

Figure: (a) A triangulated model of a dinosaur (b) Several triangle strips in the model The most common type of geometric model used for computer graphics describes the geometry of the object as a triangulated surface, as shown in Figure a. High-performance rendering engines have special hardware for rendering and shading triangles. This hardware is so fast that the bottleneck of rendering is simply feeding the structure of the triangulation into the hardware engine. Although each triangle can be described by specifying its three endpoints and any associated shading/normal information, an alternative representation is more efficient. Instead of specifying each triangle in isolation, suppose that we partition the triangles into strips of adjacent triangles and walk (b). Since each triangle shares two vertices in common with its along the strip, as shown in Figure neighbors, we save the cost of retransmitting the two extra vertices and any associated normals. To make the description of the triangles unambiguous, the Silicon Graphics triangular-mesh renderer OpenGL assumes that all turns alternate from left to right (as shown in Figure

).

Figure: Partitioning a triangular mesh into strips: (a) with left-right turns (b) with the flexibility of arbitrary turns The problem of finding a small number of strips that cover each triangle in a mesh can be thought of as a graph problem, where this graph has a vertex for every triangle of the mesh, and there is an edge between every pair of vertices representing adjacent triangles. This dual graph representation of the file:///E|/BOOK/BOOK/NODE37.HTM (1 of 4) [19/1/2003 1:28:34]

War Story: Stripping Triangulations

planar subdivision representing the triangulation (see Section triangulation needed to partition it into triangle strips. Section the graph from the triangulation.

) captures all the information about the describes our experiences constructing

Once we had the dual graph available, the project could begin in earnest. We sought to partition the vertices of the dual graph into as few paths or strips as possible. Partitioning it into one path implied that we had discovered a Hamiltonian path, which by definition visits each vertex exactly once. Since finding a Hamiltonian path was NP-complete (see Section but to concentrate instead on heuristics.

), we knew not to look for an optimal algorithm,

It is always best to start with simple heuristics before trying more complicated ones, because simple might well suffice for the job. The most natural heuristic for strip cover would be to start from an arbitrary triangle and then do a left-right walk from there until the walk ends, either by hitting the boundary of the object or a previously visited triangle. This heuristic had the advantage that it would be fast and simple, although there could be no reason to suspect that it should find the smallest possible set of left-right strips for a given triangulation. A heuristic more likely to result in a small number of strips would be greedy. Greedy heuristics always try to grab the best possible thing first. In the case of the triangulation, the natural greedy heuristic would find the starting triangle that yields the longest left-right strip, and peel that one off first. Being greedy also does not guarantee you the best possible solution, since the first strip you peel off might break apart a lot of potential strips we would have wanted to use later. Still, being greedy is a good rule of thumb if you want to get rich. Since removing the longest strip would leave the fewest number of triangles for later strips, it seemed reasonable that the greedy heuristic would out-perform the naive heuristic. But how much time does it take to find the largest strip to peel off next? Let k be the length of the walk possible from an average vertex. Using the simplest possible implementation, we could walk from each of the n vertices per iteration in order to find the largest remaining strip to report in time. With the total number of strips roughly equal to n/k, this yields an -time implementation, which would be hopelessly slow on a typical model of 20,000 triangles. How could we speed this up? It seems wasteful to rewalk from each triangle after deleting a single strip. We could maintain the lengths of all the possible future strips in a data structure. However, whenever we peel off a strip, we would have to update the lengths of all the other strips that will be affected. These strips will be shortened because they walked through a triangle that now no longer exists. There are two aspects of such a data structure:

file:///E|/BOOK/BOOK/NODE37.HTM (2 of 4) [19/1/2003 1:28:34]

War Story: Stripping Triangulations

Figure: A bounded height priority queue for triangle strips ●

Priority Queue - Since we were repeatedly interested in identifying the longest possible next strip, we needed a priority queue to store the strips, ordered according to length. The next strip to peel would always be the top of the queue. Our priority queue had to permit reducing the priority of arbitrary elements of the queue whenever we updated the strip lengths to reflect what triangles were peeled away. Because all of the strip lengths were bounded by a fairly small integer (hardware constraints prevent any strip from having more than 256 vertices), we used a bounded height priority queue (shown in Figure also have worked just fine.

●

and described in Section

). An ordinary heap would

To update a queue entry associated with a triangle, we needed to be able to quickly find where it was. This meant that we also needed a ... Dictionary - For each triangle in the mesh, we needed a way to find where it was in the queue. This meant storing a pointer for each triangle. Since each triangle was defined by three integer vertex numbers, either a hash table or an array of lists of triangles would suffice. By integrating this dictionary with the priority queue, we built a data structure capable of a wider range of operations.

Although there were various other complications, such as quickly recalculating the length of the strips affected by the peeling, the key idea needed to obtain better performance was to use the priority queue. Run time improved by several orders of magnitude after employing these data structures.

file:///E|/BOOK/BOOK/NODE37.HTM (3 of 4) [19/1/2003 1:28:34]

War Story: Stripping Triangulations

Figure: A comparison of the naive versus greedy heuristics for several triangular meshes How much better did the greedy heuristic do than the naive heuristic? Consider the table in Figure . In all cases, the greedy heuristic led to a set of strips that cost less, as measured by the total size of the strips. The savings ranged from about 10% to 50%, quite remarkable, since the greatest possible improvement (going from three vertices per triangle down to one) could yield a savings of only 66.6%. After implementing the greedy heuristic with our priority queue data structure, our complete algorithm ran in time, where n is the number of triangles and k is the length of the average strip. Thus the torus, which consisted of a small number of very long strips, took longer than the jaw, even though the latter contained over three times as many triangles. There are several lessons to be gleaned from this story. First, whenever we are working with a large enough data set, only linear or close to linear algorithms (say ) are likely to be fast enough. Second, choosing the right data structure is often the key to getting the time complexity down to this point. Finally, using a greedy or somewhat smarter heuristic over the naive approach is likely to significantly improve the quality of the results. How much the improvement is likely to be can be determined only by experimentation. Our final, optimized triangle strip program is described in [ESV96].

Next: War Story: Mystery of Up: Data Structures and Sorting Previous: Bucketing Techniques Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE37.HTM (4 of 4) [19/1/2003 1:28:34]

War Story: Mystery of the Pyramids

Next: War Story: String 'em Up: Data Structures and Sorting Previous: War Story: Stripping Triangulations

War Story: Mystery of the Pyramids That look in his eyes should have warned me even before he started talking. ``We want to use a parallel supercomputer for a numerical calculation up to 1,000,000,000, but we need a faster algorithm to do it.'' I'd seen that distant look before. Eyes dulled from too much exposure to the raw horsepower of supercomputers. Machines so fast that brute force seemed to eliminate the need for clever algorithms. So it always seemed, at least until the problems got hard enough. ``I am working with a Nobel prize winner to use a computer on a famous problem in number theory. Are you familiar with Waring's problem?'' I knew some number theory [NZ80]. ``Sure. Waring's problem asks whether every integer can be expressed at least one way as the sum of at most four integer squares. For example, . It's a cute problem. I remember proving that four squares suffice in my undergraduate number theory class. Yes, it's a famous problem, sure, but one that got solved about 200 years ago.'' ``No, we are interested in a different version of Waring's problem. A pyramidal number is a number of the form , for . Thus the first several pyramidal numbers are 1, 4, 10, 20, 35, 56, 84, 120, and 165. The conjecture since 1928 is that every integer can be represented by the sum of at most five such pyramidal numbers. We want to use a supercomputer to prove this conjecture on all numbers from 1 to 1,000,000,000.'' ``Doing a billion of anything will take a serious amount of time,'' I warned. ``The time you spend to compute the minimum representation of each number will be critical, because you are going to do it one billion times. Have you thought about what kind of an algorithm you are going to use?'' ``We have already written our program and run it on a parallel supercomputer. It works very fast on file:///E|/BOOK/BOOK/NODE38.HTM (1 of 4) [19/1/2003 1:28:35]

War Story: Mystery of the Pyramids

smaller numbers. Still, it take much too much time as soon as we get to 100,000 or so.'' ``Terrific,'' I thought. Our supercomputer junkie had discovered asymptotic growth. No doubt his algorithm ran in something like quadratic time, and he got burned as soon as n got large. ``We need a faster program in order to get to one billion. Can you help us? Of course, we can run it on our parallel supercomputer when you are ready.'' I'll confess that I am a sucker for this kind of challenge, finding better algorithms to speed up programs. I agreed to think about it and got down to work. I started by looking at the program that the other guy had written. He had built an array of all the pyramidal numbers from 1 to n. To test each number k in this range, he did a brute force test to establish whether it was the sum of two pyramidal numbers. If not, the program tested whether it was the sum of three of them, then four, and finally five of them, until it first got an answer. About 45% of the integers are expressible as the sum of three pyramidal numbers, while most of the remaining 55% require the sum of four; usually each can be represented many different ways. Only 241 integers are known to require the sum of five pyramidal numbers, the largest one being 343,867. For about half of the n numbers, this algorithm presumably went through all of the three-tests and at least some of the four-tests before terminating. Thus the total time for this algorithm would be at least

time, where n=

1,000,000,000. No wonder his program cried uncle. Anything that was going to do significantly better on a problem this large had to avoid explicitly testing all triples. For each value of k, we were seeking the smallest number of pyramidal numbers that sum . In our exactly to k. This problem has a name, the knapsack problem, and it is discussed in Section case of interest, the weights are the set of pyramidal numbers no greater than n, with an additional constraint that the knapsack holds exactly k items. The standard approach to solving knapsack precomputes the sum of smaller subsets of the items for use in computing larger subsets. In other words, if we want to know whether k is expressible as the sum of three numbers, and we have a table of all sums of two numbers, all we have to do is ask whether our number is expressible as the sum of a single number plus a number in this two-table. Thus what I needed was a table of all integers less than n that could be expressed as the sum of two of the 1,818 pyramidal numbers less than 1,000,000,000. There could be at most = 3,305,124 of them. Actually, there would be only about half this many because we could eliminate duplicates, or any sum bigger than our target. Building a data structure, such as a sorted array, to hold these numbers would be no big deal. Call this data structure the two-table. To find the minimum decomposition for a given k, I would start out by checking whether it was one of

file:///E|/BOOK/BOOK/NODE38.HTM (2 of 4) [19/1/2003 1:28:35]

War Story: Mystery of the Pyramids

the 1,818 pyramidal numbers. If not, I would then search to see whether k was in the sorted table of the sums of two pyramidal numbers. If it wasn't, to see whether k was expressible as the sum of three such . This could be numbers, all I had to do was check whether k - p[i] was in the two-table for done quickly using binary search. To see whether k was expressible as the sum of four pyramidal . However, since numbers, I had to check whether k - two[i] was in the two-table for all almost every k was expressible in many ways as the sum of four pyramidal numbers, this latter test would terminate quickly, and the time taken would be dominated by the cost of the threes. Testing . Running this on each of the n whether k was the sum of three pyramidal numbers would take integers gives an algorithm for the complete job. Comparing this to his algorithm for n= 1,000,000,000 suggested that my algorithm was a cool 30,000 times faster than the original! My first attempt to code this up solved up to n=1,000,000 on my crummy Sparc ELC in about 20 minutes. From here, I experimented with different data structures to represent the sets of numbers and different algorithms to search these tables. I tried using hash tables and bit vectors instead of sorted ). My arrays and experimented with variants of binary search such as interpolation search (see Section reward for this work was solving up to n=1,000,000 in under three minutes, a factor of six improvement over my original program. With the real thinking done, I worked to tweak a little more performance out of the program. I avoided doing a sum-of-four computation on any k when k-1 was the sum-of-three, since 1 is a pyramidal number, saving about 10% of the total run time on this trick alone. Finally, I got out my profiler and tried some low-level tricks to squeeze a little more performance out of the code. For example, I saved another 10% by replacing a single procedure call with in-line code. At this point, I turned the code over to the supercomputer guy. What he did with it is a depressing tale, which is reported in Section

.

In writing up this war story, I went back to rerun our program almost five years later. On my desktop Sparc 5, getting to 1,000,000 now took 167.5 seconds using the cc compiler without turning on any compiler optimization. With level 2 optimization, the job ran in only 81.8 seconds, quite a tribute to the quality of the optimizer. The gcc compiler with optimization did even better, taking only 74.9 seconds to get to 1,000,000. The run time on my desktop machine had improved by a factor of about three over a four-year period. This is probably typical for most desktops. The primary importance of this war story is to illustrate the enormous potential for algorithmic speedups, as opposed to the fairly limited speedup obtainable via more expensive hardware. I sped his program up by about 30,000 times. His million-dollar computer had 16 processors, each reportedly five times faster on integer computations than the $3,000 machine on my desk. That gave a maximum potential speedup of less than 100 times. Clearly, the algorithmic improvement was the big winner here, as it is certain to be in any sufficiently large computation.

file:///E|/BOOK/BOOK/NODE38.HTM (3 of 4) [19/1/2003 1:28:35]

War Story: Mystery of the Pyramids

Next: War Story: String 'em Up: Data Structures and Sorting Previous: War Story: Stripping Triangulations Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE38.HTM (4 of 4) [19/1/2003 1:28:35]

War Story: String 'em Up

Next: Exercises Up: Data Structures and Sorting Previous: War Story: Mystery of

War Story: String 'em Up Biologists are hard at work on a fifteen-year project to sequence the human genome. This pattern of nucleotides encodes all the information necessary to build the proteins that we are built of. This project has already had an enormous impact on medicine and molecular biology. Algorists have become interested in the human genome project as well, for several reasons: ●

DNA sequences can be accurately represented as strings of characters on the four-letter alphabet (A,C,T,G). Biologist's needs have sparked new interest in old algorithmic problems (such as string matching - see Section

●

●

) as well as creating new problems of substantial interest (such as

). shortest common superstring - see Section DNA sequences are very long strings. The human genome is approximately three billion base pairs (or characters) long. Thus sophisticated computational techniques are necessary to deal with them. Such large problem sizes means that asymptotic (big-Oh) complexity analysis is usually fully justified on biological problems. Enough money is being invested in the human genome project for computer scientists to want to claim their piece of the action.

My particular interest in computational biology has revolved around a recently proposed but algorithmically intensive technique for DNA sequencing called sequencing by hybridization (SBH) [CK94, PL94]. The traditional sequencing by hybridization procedure attaches a set of probes to an array, forming a sequencing chip. Each of these probes determines whether or not the probe string occurs as a substring of the DNA target. The target DNA can now be sequenced based on the constraints of which strings are and are not substrings of the target. strings) are necessary to sequence One problem with SBH is that enormous arrays (say relatively short pieces of DNA (typically about 200 base pairs long). The major reason is that all of these probes are made at the same time. If you want to look up a name in the telephone book but are only allowed to consult the book once, you must copy down every single name from the book at that time. But if you are allowed to ask ``is the name before Mendoza?'' and wait for the answer before asking your next question, you can use binary search to greatly reduce your total effort. file:///E|/BOOK/BOOK/NODE39.HTM (1 of 5) [19/1/2003 1:28:38]

War Story: String 'em Up

We were convinced that using several small arrays would be more efficient than using one big array. We even had theory to justify our technique, but biologists aren't very inclined to believe theory. They demand experiments for proof. Hence we had to implement our algorithms and use simulation to prove that they worked. So much for motivation. The rest of this tale will demonstrate the impact that clever data structures can have on a string processing application. Our technique involved identifying all the strings of length 2k that are possible substrings of an unknown string S, given that we know all length k substrings of S. For example, suppose we know that AC, CA, and CC are the only length-2 substrings of S. It is certainly possible that ACCA is a substring of S, since the center substring is one of our possibilities. However, CAAC cannot be a substring of S, since we know that AA is not a substring of S. We needed to find a fast algorithm to construct all the consistent length-2k strings, since S could be very long.

Figure: The concatentation of two fragments can be in S only if all subfragments are The simplest algorithm to build the 2k strings would be to concatenate all pairs of k-strings together, and then for each pair to make sure that all (k-1) length-k substrings spanning the boundary of . For example, the nine possible the concatenation were in fact substrings, as shown in Figure concatenations of AC, CA, and CC are ACAC, ACCA, ACCC, CAAC, CACA, CACC, CCAC, CCCA, and CCCC. Only CAAC can be eliminated because of the absence of AA. We needed a fast way of testing whether each of the k-1 substrings straddling the concatenation was a member of our dictionary of permissible k-strings. The time it takes to do this depends upon which kind of data structure we use to maintain this dictionary. With a binary search tree, we could find the correct string within comparisons, where each comparison involved testing which of two length-k strings appeared first in alphabetical order. Since each such comparison could require testing k pairs of characters, the total time using a binary search tree would be . That seemed pretty good. So my graduate student Dimitris Margaritis implemented a binary search tree data structure for our implementation. It worked great up until the moment we ran it.

file:///E|/BOOK/BOOK/NODE39.HTM (2 of 5) [19/1/2003 1:28:38]

War Story: String 'em Up

``I've tried the fastest computer in our department, but our program is too slow,'' Dimitris complained. ``It takes forever on strings of length only 2,000 characters. We will never get up to 50,000.'' For interactive SBH to be competitive as a sequencing method, we had to be able to sequence long fragments of DNA, ideally over 50 kilobases in length. If we couldn't speed up the program, we would be in the embarrassing position of having a biological technique invented by computer scientists fail because the computations took too long. We profiled our program and discovered that almost all the time was spent searching in this data structure, which was no surprise. For each of the possible concatenations, we did this k-1 times. We needed a faster dictionary data structure, since search was the innermost operation in such a deep loop. ``What about using a hash table?'' I suggested. ``If we do it right, it should take O(k) time to hash a kcharacter string and look it up in our table. That should knock off a factor of , which will mean something when 2,000.'' Dimitris went back and implemented a hash table implementation for our dictionary. Again, it worked great up until the moment we ran it. ``Our program is still too slow,'' Dimitris complained. ``Sure, it is now about ten times faster on strings of length 2,000. So now we can get up to about 4,000 characters. Big deal. We will never get up to 50,000.'' ``We should have expected only a factor ten speedup,'' I mused. ``After all, faster data structure to search in our dictionary of strings.''

. We need a

``But what can be faster than a hash table?'' Dimitris countered. ``To look up a k-character string, you must read all k characters. Our hash table already does O(k) searching.'' ``Sure, it takes k comparisons to test the first substring. But maybe we can do better on the second test. Remember where our dictionary queries are coming from. When we concatenate ABCD with EFGH, we are first testing whether BCDE is in the dictionary, then CDEF. These strings differ from each other by only one character. We should be able to exploit this so that each subsequent test takes constant time to perform....'' ``We can't do that with a hash table,'' Dimitris observed. ``The second key is not going to be anywhere near the first in the table. A binary search tree won't help, either. Since the keys ABCD and BCDE differ according to the first character, the two strings will be in different parts of the tree.''

file:///E|/BOOK/BOOK/NODE39.HTM (3 of 5) [19/1/2003 1:28:38]

War Story: String 'em Up

Figure: Suffix tree on ACAC and CACT, with the pointer to the suffix of ACAC ``But we can use a suffix tree to do this,'' I countered. ``A suffix tree is a trie containing all the suffixes of a given set of strings. For example, the suffixes of ACAC are . Coupled with suffixes of string CACT, we get the suffix tree of Figure . By following a pointer from ACAC to its longest proper suffix CAC, we get to the right place to test whether CACT is in our set of strings. One character comparison is all we need to do from there.'' Suffix trees are amazing data structures, discussed in considerably more detail in Section . Dimitris did some reading about them, then built a nice suffix tree implementation for our dictionary. Once again, it worked great up until the moment we ran it. ``Now our program is faster, but it runs out of memory,'' Dimitris complained. ``And this on a 128 megabyte machine with 400 megabytes virtual memory! The suffix tree builds a path of length k for each suffix of length k, so all told there can be nodes in the tree. It crashes when we go beyond 2,000 characters. We will never get up to strings with 50,000 characters.'' I wasn't yet ready to give up. ``There is a way around the space problem, by using compressed suffix trees,'' I recalled. ``Instead of explicitly representing long paths of character nodes, we can refer back to the original string.'' Compressed suffix trees always take linear space, as described in Section

.

Dimitris went back one last time and implemented the compressed suffix tree data structure. Now it , we ran our simulation for strings of length n= 65,536 on a worked great! As shown in Figure SPARCstation 20 without incident. Our results, reported in [MS95a], showed that interactive SBH could be a very efficient sequencing technique. Based on these simulations, we were able to arouse interest in our technique from biologists. Making the actual wet laboratory experiments feasible provided another computational challenge, which is reported in Section

file:///E|/BOOK/BOOK/NODE39.HTM (4 of 5) [19/1/2003 1:28:38]

.

War Story: String 'em Up

The take home lessons for programmers from Figure should be apparent. We isolated a single operation (dictionary string search) that was being performed repeatedly and optimized the data structure we used to support it. We started with a simple implementation (binary search trees) in the hopes that it would suffice, and then used profiling to reveal the trouble when it didn't. When an improved dictionary structure still did not suffice, we looked deeper into what kinds of queries we were performing, so that we could identify an even better data structure. Finally, we didn't give up until we had achieved the level of performance we needed. In algorithms, as in life, persistence usually pays off.

Figure: Run times (in seconds) for the SBH simulation using various data structures

Next: Exercises Up: Data Structures and Sorting Previous: War Story: Mystery of Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE39.HTM (5 of 5) [19/1/2003 1:28:38]

Exercises

Next: Implementation Challenges Up: Data Structures and Sorting Previous: War Story: String 'em

Exercises 1. Newt Gingrich is given the job of partitioning 2n players into two teams of n players each. Each player has a numerical rating that measures how good he/she is at the game. Newt seeks to divide the players as unfairly as possible, so as to create the biggest possible talent imbalance between team A and team B. Show how Newt can do the job in time. 2. Take as input a sequence of 2n real numbers. Design an algorithm that partitions the numbers into n pairs, with the property that the partition minimizes the maximum sum of a pair. For example, say we are given the numbers (1,3,5,9). The possible partitions are ((1,3),(5,9)), ((1,5),(3,9)), and ((1,9),(3,5)). The pair sums for these partitions are (4,14), (6,12), and (10,8). Thus the third partition has 10 as its maximum sum, which is the minimum over the three partitions. 3. Assume that we are given as input n pairs of items, where the first item is a number and the second item is one of three colors (red, blue, or yellow). Further, assume that the items are sorted by number. Give an O(n) algorithm to sort the items by color (all reds before all blues before all yellows) such that the numbers for identical colors stay sorted. For example: (1,blue), (3,red), (4,blue), (6,yellow), (9,red) should become (3,red), (9,red), (1,blue), (4,blue), (6,yellow). 4. (*) The mode of a set of numbers is the number that occurs most frequently in the set. The set (4,6,2,4,3,1) has a mode of 4. 1. Give an efficient and correct algorithm to compute the mode of a set of n numbers. 2. Suppose we know that there is an (unknown) element that occurs n/2+1 times in the set. Give a worst-case linear-time algorithm to find the mode. For partial credit, your algorithm may run in expected linear time. 5. Given two sets and (each of size n), and a number x, describe an algorithm for finding whether there exists a pair of elements, one from and one from , that add up to x. algorithm for this problem.) (For partial credit, give a 6. For each of the following problems, give an algorithm that finds the desired numbers within the given amount of time. To keep your answers brief, feel free to use algorithms from the book as , 19-3 maximizes the difference, while 8-6 subroutines. For the example, minimizes the difference.

file:///E|/BOOK/BOOK/NODE40.HTM (1 of 4) [19/1/2003 1:28:40]

Exercises

(a) Let S be an unsorted array of n integers. Give an algorithm that finds the pair maximizes |x-y|. Your algorithm must run in O(n) worst-case time. (b) Let S be a sorted array of n integers. Give an algorithm that finds the pair maximizes |x-y|. Your algorithm must run in O(1) worst-case time. (c) Let S be an unsorted array of n integers. Give an algorithm that finds the pair minimizes |x-y|, for . Your algorithm must run in worst-case time.

7.

8.

9.

10.

11. 12.

that

that

that

that (d) Let S be a sorted array of n integers. Give an algorithm that finds the pair minimizes |x-y|, for . Your algorithm must run in O(n) worst-case time. (*) Describe how to modify any balanced tree data structure such that search, insert, delete, time each, but successor and predecessor now take O(1) minimum, and maximum still take time each. Which operations have to be modified to support this? (*) In one of my research papers [Ski88], I give a comparison-based sorting algorithm that runs in . Given the existence of an lower bound for sorting, how can this be possible? (*) Suppose you have access to a balanced dictionary data structure, which supports each of the time. operations search, insert, delete, minimum, maximum, successor, and predecessor in Explain how to modify the insert and delete operations so they still take but now minimum and maximum take O(1) time. (Hint: think in terms of using the abstract dictionary operations, instead of mucking about with pointers and the like.) (*) Mr. B. C. Dull claims to have developed a new data structure for priority queues that supports the operations Insert, Maximum, and Extract-Max, all in O(1) worst-case time. Prove that he is mistaken. (Hint: the argument does not involve a lot of gory details-just think about what this would imply about the lower bound for sorting.) Use the partitioning step of Quicksort to give an algorithm that finds the median element of an array of n integers in expected O(n) time. (*) You are given the task of reading in n numbers and then printing them out in sorted order. Suppose you have access to a balanced dictionary data structure, which supports each of the operations search, insert, delete, minimum, maximum, successor, and predecessor in time. ❍

❍

❍

Explain how you can use this dictionary to sort in abstract operations: minimum, successor, insert, search. Explain how you can use this dictionary to sort in abstract operations: minimum, insert, delete, search. Explain how you can use this dictionary to sort in abstract operations: insert and in-order traversal.

time using only the following time using only the following time using only the following

13. The running time for Quicksort depends upon both the data being sorted and the partition rule

file:///E|/BOOK/BOOK/NODE40.HTM (2 of 4) [19/1/2003 1:28:40]

Exercises

used to select the pivot. Although Quicksort is on average, certain partition rules cause Quicksort to take time if the array is already sorted. (a) Suppose we always pick the pivot element to be the key from the last position of the subarray. On a sorted array, does Quicksort now take , , or ? (b) Suppose we always pick the pivot element to be the key from the middle position of the subarray. On a sorted array, does Quicksort now take , , or ? (c) Suppose we always pick the pivot element to be the key of the median element of the first three keys of the subarray. (The median of three keys is the middle value, so the median of 5, 3, 8 is five.) On a sorted array, does Quicksort now take , , or ?

14. 15.

16.

17.

(d) Suppose we always pick the pivot element to be the key of the median element of the first, last, and middle elements of the subarray. On a sorted array, does Quicksort now take , , or ? algorithm to test whether k of (*) Given a set S of n integers and an integer T, give an the integers in S sum up to T. (**) Design a data structure that allows one to search, insert, and delete an integer X in O(1) time and (i.e. constant time, independent of the total number of integers stored). Assume that that there are m+n units of space available for the symbol table, where m is the maximum number of integers that can be in the table at any one time. (Hint: use two arrays A[1..n] and B[1..m].) You are not allowed to initialize either A or B, as that would take O(m) or O(n) operations. This means the arrays are full of random garbage to begin with, so you must be very careful. (*) Let P be a simple, but not necessarily convex, polygon and q an arbitrary point not necessarily in P. Design an efficient algorithm to find a line segment originating from q that intersects the maximum number of edges of P. In other words, if standing at point q, in what direction should you aim a gun so the bullet will go through the largest number of walls. A bullet through a vertex algorithm is possible. of P gets credit for only one wall. An (**) The onion of a set of n points is the series of convex polygons that result from finding the convex hull, striping it from the point set, and repeating until no more points are left. Give an algorithm for determining the onion of a point set.

Next: Implementation Challenges Up: Data Structures and Sorting Previous: War Story: String 'em Algorithms file:///E|/BOOK/BOOK/NODE40.HTM (3 of 4) [19/1/2003 1:28:40]

Exercises

Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE40.HTM (4 of 4) [19/1/2003 1:28:40]

Dynamic Programming

Next: Fibonacci numbers Up: Breaking Problems Down Previous: Breaking Problems Down

Dynamic Programming After you understand it, dynamic programming is probably the easiest algorithm design technique to apply in practice. In fact, I find that dynamic programming algorithms are usually easier to reinvent than to try to look up in a book. Until you understand it, however, dynamic programming seems like magic. You have to figure out the trick before you can use it. In algorithms for problems such as sorting, correctness tends to be easier to verify than efficiency. This is not the case for optimization problems, where we seek to find a solution that maximizes or minimizes some function. In designing algorithms for an optimization problem, we must prove that our algorithm always gives the best possible solution. Greedy algorithms, which make the best local decision at each step, occasionally happen to produce a global optimum for certain problems. These are typically efficient. However, you need a proof to show that you always end up with the best answer. Exhaustive search algorithms, which try all possibilities and select the best, by definition must always produce the optimum result, but usually at a prohibitive cost in terms of time complexity. Dynamic programming combines the best of both worlds. The technique systematically considers all possible decisions and always selects the one that proves to be the best. By storing the consequences of all possible decisions to date and using this information in a systematic way, the total amount of work is minimized. Dynamic programming is best learned by carefully studying a number of examples until things start to click.

● ● ● ● ●

Fibonacci numbers The Partition Problem Approximate String Matching Longest Increasing Sequence Minimum Weight Triangulation

file:///E|/BOOK/BOOK2/NODE43.HTM (1 of 2) [19/1/2003 1:28:40]

Dynamic Programming

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE43.HTM (2 of 2) [19/1/2003 1:28:40]

Fibonacci numbers

Next: The Partition Problem Up: Dynamic Programming Previous: Dynamic Programming

Fibonacci numbers The tradeoff between space and time exploited in dynamic programming is best illustrated in evaluating recurrence relations, such as the Fibonacci numbers. The Fibonacci numbers were originally defined by the Italian mathematician Fibonacci in the thirteenth century to model the growth of rabbit populations. Rabbits breed, well, like rabbits. Fibonacci surmised that the number of pairs of rabbits born in a given year is equal to the number of pairs of rabbits born in each of the two previous years, if you start with one pair of rabbits in the first year. To count the number of rabbits born in the nth year, he defined the following recurrence relation:

with basis cases

and

. Thus , , and the series continues . As it turns out, Fibonacci's formula didn't do a very good job of counting rabbits, but it does have a host of other applications and interesting properties. Since they are defined by a recursive formula, it is easy to write a recursive program to compute the nth Fibonacci number. Most students have to do this in one of their first programming courses. Indeed, I have particularly fond memories of pulling my hair out writing such a program in 8080 assembly language. In pseudocode, the recursive algorithm looks like this:

Fibonacci[n]

if (n=0) then return(0)

else if (n=1) then return(1)

else return(Fibonacci[n-1]+Fibonacci[n-2])

file:///E|/BOOK/BOOK2/NODE44.HTM (1 of 3) [19/1/2003 1:28:42]

Fibonacci numbers

Figure: The computation tree for computing Fibonacci numbers recursively How much time does this algorithm take to compute Fibonacci[n]? Since , this means that . Since our recursion tree, illustrated in , has only 0 and 1 as leaves, summing up to such a large number means we must have at least Figure leaves or procedure calls! This humble little program takes exponential time to run! In fact, we can do much better. We can calculate for time in the algorithm below:

in linear time by storing all values. We trade space

Fibonacci[n]

For i=1 to n, Because we evaluate the Fibonacci numbers from smallest to biggest and store all the results, we know and ready whenever we need to compute . Thus each of the n values is computed that we have as the simple sum of two integers in total O(n) time, which is quite an improvement over exponential time.

file:///E|/BOOK/BOOK2/NODE44.HTM (2 of 3) [19/1/2003 1:28:42]

Fibonacci numbers

Next: The Partition Problem Up: Dynamic Programming Previous: Dynamic Programming Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE44.HTM (3 of 3) [19/1/2003 1:28:42]

The Partition Problem

Next: Approximate String Matching Up: Dynamic Programming Previous: Fibonacci numbers

The Partition Problem

Suppose that three workers are given the task of scanning through a shelf of books in search of a given piece of information. To get the job done fairly and efficiently, the books are to be partitioned among the three workers. To avoid the need to rearrange the books or separate them into piles, it would be simplest to divide the shelf into three regions and assign each region to one worker. But what is the fairest way to divide the shelf up? If each book is the same length, say 100 pages, the job is pretty easy. Just partition the books into equal-sized regions,

so that everyone has 300 pages to deal with. But what if the books are not the same length? Suppose we used the same partition when the book sizes looked like this:

I, for one, would volunteer to take the first section, with only 600 pages to scan, instead of the last one, with 2,400 pages. The fairest possible partition for this shelf would be

where the largest job is only 1,700 pages and the smallest job 1,300. In general, we have the following problem: Input: A given arrangement S of non-negative numbers

and an integer k.

Output: Partition S into k ranges, so as to minimize the maximum sum over all the ranges. This so-called linear partition problem arises often in parallel processing, since we seek to balance the work done across processors so as to minimize the total elapsed run time. Indeed, the war story of Section

revolves around a botched solution to this problem.

Stop for a few minutes and try to find an algorithm to solve the linear partition problem.

The beginning algorist might suggest a heuristic as the most natural approach to solve the partition problem. Perhaps they would compute the average size of a partition, , and then try to insert the dividers so as to come close to this average. However, such heuristic methods are doomed to fail on certain inputs, because they do not systematically evaluate all possibilities. Instead, consider a recursive, exhaustive search approach to solving this problem. Notice that the kth partition starts right after we placed the (k-1)st divider. Where can we place this last divider? Between the ith and (i+1)st elements for some i, where file:///E|/BOOK/BOOK2/NODE45.HTM (1 of 5) [19/1/2003 1:28:44]

The Partition Problem

. What is the cost of this? The total cost will be the larger of two quantities, (1) the cost of the last partition and (2) the cost of the largest partition cost formed to the left of i. What is the size of this left partition? To minimize our total, as equally as possible. This is a smaller we would want to use the k-2 remaining dividers to partition the elements instance of the same problem, and hence can be solved recursively! Therefore, let us define M[n,k] to be the minimum possible cost over all partitionings of into k ranges, where the cost of a partition is the largest sum of elements in one of its parts. Thus defined, this function can be evaluated:

with the natural basis cases of

By definition, this recurrence must return the size of the optimal partition. But how long does it take? If we evaluate the recurrence without storing partial results, we will be doomed to spend exponential time, repeatedly recalculating the same quantities. However, by storing the computed values and looking them up as needed, we can save an enormous amount of time. How long does it take to compute this when we store the partial results? Well, how many results are computed? A total of cells exist in the table. How much time does it take to compute the result M[n',k']? Well, calculating this quantity involves finding the minimum of n' quantities, each of which is the maximum of the table lookup and a sum of at most n' elements. If filling each of k n boxes takes at most time per box, the total recurrence can be computed in time. To complete the implementation, we must specify the boundary conditions of the recurrence relation and an order to evaluate it in. These boundary conditions always settle the smallest possible values for each of the arguments of the recurrence. For this problem, the smallest reasonable value of the first argument is n=1, meaning that the first partition consists of a single element. We can't create a first partition smaller than regardless of how many dividers are used. The smallest reasonable value of the second argument is k=1, implying that we do not partition S at all. The evaluation order computes the smaller values before the bigger values, so that each evaluation has what it needs waiting for it. Full details are provided in the pseudocode below:

Partition[S,k]

(* compute prefix sums:

*)

p[0] = 0

for i=1 to n do

(* initialize boundary conditions *) file:///E|/BOOK/BOOK2/NODE45.HTM (2 of 5) [19/1/2003 1:28:44]

The Partition Problem

for i=1 to n do M[i,1] = p[i]

for i=1 to k do

(* evaluate main recurrence *)

for i=2 to n do

for j = 2 to k do

for x = 1 to i-1 do

if (M[i,j] > ) then

M[i,j] =

D[i,j] = x

Figure: Dynamic programming matrices M and D in partitioning

file:///E|/BOOK/BOOK2/NODE45.HTM (3 of 5) [19/1/2003 1:28:44]

The Partition Problem

Figure: Dynamic programming matrices M and D in partitioning The implementation above in fact runs faster than advertised. Our original analysis assumed that it took time to update each cell of the matrix. This is because we selected the best of up to n possible points to place the divider, each of which requires the sum of up to n possible terms. In fact, it is easy to avoid the need to compute these sums by storing the set of n , since . This enables us to evaluate the recurrence in linear time per cell, yielding prefix sums an

algorithm.

and , you should be able to By studying the recurrence relation and the dynamic programming matrices of Figures convince yourself that the final value of M(n,k) will be the cost of the largest range in the optimal partition. However, what good is that? For most applications, what we need is the actual partition that does the job. Without it, all we are left with is a coupon for a great price on an out-of-stock item. The second matrix, D, is used to reconstruct the optimal partition. Whenever we update the value of M[i,j], we record which divider position was required to achieve that value. To reconstruct the path used to get to the optimal solution, we work backward from D[n,k] and add a divider at each specified position. This backwards walking is best achieved by a recursive subroutine:

ReconstructPartition(S,D,n,k)

If (k = 1) then print the first partition

else

ReconstructPartition(S,D,D[n,k],k-1)

Print the kth partition {

}

It is important to catch the distinction between storing the value of a cell and what decision/move it took to get there. The latter is not used in the computation but is presumably the real thing that you are interested in. For most of the examples in this chapter, we will not worry about reconstructing the answer. However, study this example closely to ensure that you know how to obtain the winning configuration when you need it.

file:///E|/BOOK/BOOK2/NODE45.HTM (4 of 5) [19/1/2003 1:28:44]

The Partition Problem

Next: Approximate String Matching Up: Dynamic Programming Previous: Fibonacci numbers Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE45.HTM (5 of 5) [19/1/2003 1:28:44]

Approximate String Matching

Next: Longest Increasing Sequence Up: Dynamic Programming Previous: The Partition Problem

Approximate String Matching

An important task in text processing is string matching - finding all the occurrences of a word in the text. Unfortunately, many words in documents are mispelled (sic). How can we search for the string closest to a given pattern in order to account for spelling errors? To be more precise, let P be a pattern string and T a text string over the same alphabet. The edit distance between P and T is the smallest number of changes sufficient to transform a substring of T into P, where the changes may be: 1. Substitution - two corresponding characters may differ: KAT CAT. 2. Insertion - we may add a character to T that is in P: CT CAT. 3. Deletion - we may delete from T a character that is not in P: CAAT CAT. For example, P=abcdefghijkl can be matched to T=bcdeffghixkl using exactly three changes, one of each of the above types. Approximate string matching arises in many applications, as discussed in Section . It seems like a difficult problem, because we have to decide where to delete and insert characters in pattern and text. But let us think about the problem in reverse. What information would we like to have in order to make the final decision; i.e. what should happen with the last character in each string? The last characters may be either be matched, if they are identical, or otherwise substituted one for the other. The only other options are inserting or deleting a character to take care of the last character of either the pattern or the text. More precisely, let D[i,j] be the minimum number of differences between D[i, j] is the minimum of the three possible ways to extend smaller strings:

and the segment of T ending at j.

, then D[i-1, j-1], else D[i-1, j-1]+1. This means we either match or substitute the ith and jth characters, 1. If depending upon whether they do or do not match. 2. D[i-1, j]+1. This means that there is an extra character in the pattern to account for, so we do not advance the text pointer and pay the cost of an insertion. 3. D[i, j-1]+1. This means that there is an extra character in the text to remove, so we do not advance the pattern pointer and pay the cost of a deletion. The alert reader will notice that we have not specified the boundary conditions of this recurrence relation. It is critical to get the initialization right if our program is to return the correct edit distance. The value of D[0,i] will correspond to the cost of matching the first i characters of the text with none of the pattern. What this value should be depends upon what you want to compute. If you seek to match the entire pattern against the entire text, this means that we must delete the first i characters of the text, so D[0,i] = i to pay the cost of the deletions. But what if we want to find where the pattern occurs in a long text? It should not cost more if the matched pattern starts far into the text than if it is near the front. Therefore, the starting cost should be equal for all positions. In this case, D[0,i] = 0, since we pay no cost for deleting the first i characters of the text. In both cases, D[i,0] = i, since we cannot excuse deleting the first i characters of the pattern without penalty.

file:///E|/BOOK/BOOK2/NODE46.HTM (1 of 3) [19/1/2003 1:28:46]

Approximate String Matching

Figure: Example dynamic programming matrix for edit distance computation, with the optimal alignment path highlighted in bold Once you accept the recurrence, it is straightforward to turn it into a dynamic programming algorithm that creates an matrix D, where n = |P| and m = |T|. Here it is, initialized for full pattern matching:

EditDistance(P,T)

(*initialization*)

For i = 0 to n do D[i,0] = i

For i = 0 to m do D[0,i] = i

(*recurrence*)

For i = 1 to n do

For j = 1 to m do

D[i-1,j]+1, D[i,j1]+1 ) How much time does this take? To fill in cell D[i,j], we need only compare two characters and look at three other cells. Since it requires only constant time to update each cell, the total time is O(mn).

file:///E|/BOOK/BOOK2/NODE46.HTM (2 of 3) [19/1/2003 1:28:46]

Approximate String Matching

The value to return as the answer to our pattern matching problem depends on what we are interested in. If we only needed the cost of comparing all of the pattern against all of the text, such as in comparing the spelling of two words, all we would . But what if we need to identify the best matching substring in the text? need is the cost of D[n,m], as shown in Figure Assuming that the initialization was performed correctly for such substring matching, we seek the cheapest matching of the full pattern ending anywhere in the text. This means the cost equals , i.e. the smallest cost on the last row of D. Of course, this only gives the cost of the optimal matching, while we are often interested in reconstructing the actual alignment - which characters got matched, substituted, and deleted. These can be reconstructed from the pattern/text and table without an auxiliary storage, once we have identified the cell with the lowest cost. From this cell, we want to walk upwards and backwards through the matrix. Given the costs of its three neighbors and the corresponding characters, we can reconstruct which choice was made to get to the goal cell. The direction of each backwards step (to the left, up, or diagonal to the upper left) identifies whether it was an insertion, deletion, or match/substitution. Ties can be broken arbitrarily, since either way costs the same. We keep walking backwards until we hit the end of the matrix, specifying the starting point. This backwards-walking phase takes O(n+m) time, since we traverse only the cells involved in the alignment. The alert reader will notice that it is unnecessary to keep all O(mn) cells to compute the cost of an alignment. If we evaluate the recurrence by filling in the columns of the matrix from left to right, we will never need more than two columns of cells to store what is necessary for the computation. Thus O(m) space is sufficient to evaluate the recurrence without changing the time complexity at all. Unfortunately, without the full matrix we cannot reconstruct the alignment. Saving space in dynamic programming is very important. Since memory on any computer is limited, O(nm) space proves more of a bottleneck than O(nm) time. Fortunately, there is a clever divide-and-conquer algorithm that computes the actual alignment in O(nm) time and O(m) space. This algorithm is discussed in Section

.

Next: Longest Increasing Sequence Up: Dynamic Programming Previous: The Partition Problem Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE46.HTM (3 of 3) [19/1/2003 1:28:46]

Longest Increasing Sequence

Next: Minimum Weight Triangulation Up: Dynamic Programming Previous: Approximate String Matching

Longest Increasing Sequence

Hopefully, a pattern is emerging. Every dynamic programming solution has three components: 1. Formulate the answer as a recurrence relation or recursive algorithm. 2. Show that the number of different values of your recurrence is bounded by a (hopefully small) polynomial. 3. Specify an order of evaluation for the recurrence so you always have the partial results you need available when you need them. To see how this is done, let's see how we would develop an algorithm to find the longest monotonically increasing sequence in a sequence of n numbers. This problem arises in pattern matching on permutations, as described in Section . We distinguish an increasing sequence from a run, in that the selected elements need not be neighbors of each other. The selected elements must be in sorted order from left to right. For example, consider the sequence

The longest increasing subsequence of has length 3 and is either (2,3,4) or (2,3,6). The longest increasing run is of length 2, either (2,8) or (1,6). Finding the longest increasing run in a numerical sequence is straightforward, indeed you should be able to devise a linear-time algorithm fairly easily. However, finding the longest increasing subsequence is considerably trickier. How can we identify which scattered elements to skip? To apply dynamic programming, we need to construct a recurrence computing the length of the longest sequence. To find the right recurrence, ask what information about the first n-1 elements of S, coupled with the last element , would enable you to find the answer for the entire sequence? ●

The length of the longest increasing sequence in seems a useful thing to know. In fact, this will be the longest increasing sequence in S, unless extends it or another increasing sequence of the same length. Unfortunately, knowing just this length is not enough. Suppose I told you that the longest increasing sequence in was of length 5 and that .

file:///E|/BOOK/BOOK2/NODE47.HTM (1 of 3) [19/1/2003 1:28:48]

Longest Increasing Sequence

●

Will the length of the final longest increasing subsequence of S be 5 or 6? Therefore, we also need the length of the longest sequence that will extend. To be certain we know this, we really need the length of the longest sequence that any possible number can extend.

This provides the idea around which to build a recurrence. Define to be the length of the longest sequence ending with . Verify that the following table is correct: sequence

9 5 2 8 7 3 1 6 4

length

1 1 1 2 2 2 1 3 3

predecessor

- - - 2 2 3 - 6 6

The longest increasing sequence containing the nth number will be formed by appending it to the longest increasing sequence to the left of n that ends on a number smaller than . The following recurrence computes :

These values define the length of the longest increasing sequence ending at each number. The length of the longest increasing subsequence of the entire permutation is given by , since the winning sequence will have to end somewhere. What is the time complexity of this algorithm? Each one of the n values of is computed by comparing against up to values to the left of it, so this analysis gives a total of time. In fact, by time. using dictionary data structures in a clever way, we can evaluate this recurrence in However, the simple recurrence would be easy to program and therefore is a good place to start. What auxiliary information will we need to store in order to reconstruct the actual sequence instead of its length? For each element , we will store the index of the element that appears immediately before in the longest increasing sequence ending at . Since all of these pointers go towards the left, it is a simple matter to start from the last value of the longest sequence and follow the pointers so as to reconstruct the other items in the sequence.

Next: Minimum Weight Triangulation Up: Dynamic Programming Previous: Approximate String Matching file:///E|/BOOK/BOOK2/NODE47.HTM (2 of 3) [19/1/2003 1:28:48]

Longest Increasing Sequence

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE47.HTM (3 of 3) [19/1/2003 1:28:48]

Minimum Weight Triangulation

Next: Limitations of Dynamic Programming Up: Dynamic Programming Previous: Longest Increasing Sequence

Minimum Weight Triangulation

Figure: Two different triangulations of a given convex seven-gon A triangulation of a polygon

is a set of non-intersecting diagonals that partitions the polygon into

triangles. We say that the weight of a triangulation is the sum of the lengths of its diagonals. As shown in Figure , any given polygon may have many different triangulations. For any given polygon, we seek to find its minimum weight triangulation. Triangulation is a fundamental component of most geometric algorithms, as discussed in Section

.

To apply dynamic programming, we need a way to carve up the polygon into smaller pieces. A first idea might be to try all possible chords, each of which partitions the polygon into two smaller polygons. Using dynamic programming, this will work to give a polynomial-time algorithm. However, there is a slicker approach.

Figure: Selecting the vertex k to pair with an edge (i,j) of the polygon Observe that every edge of the input polygon must be involved in exactly one triangle. Turning this edge into a triangle means identifying the third vertex, as shown in Figure . Once we find the correct connecting vertex, the polygon will be partitioned into two smaller pieces, both of which need to be triangulated optimally. Let T[i,j] be the cost of triangulating from vertex to vertex , ignoring the length of the chord from to . The latter clause avoids double counting these internal chords in the following recurrence:

The basis condition applies when i and j are immediate neighbors, as T[i,i+1] = 0. Since the number of vertices in each subrange of the right side of the recurrence is smaller than that on the left side, evaluation can proceed in terms of the gap size from i to j:

Minimum-Weight-Triangulation(P)

file:///E|/BOOK/BOOK2/NODE48.HTM (1 of 2) [19/1/2003 1:28:49]

Minimum Weight Triangulation

for i=1 to n do T[i, j]=0

for gap=1 to n-1

for i=1 to n-gap do

j=i+gap

return T[1, n]

There are

values of T, each of which takes O(j-i) time if we evaluate the sections in order of increasing size. Since j-i

= O(n), complete evaluation takes

time and

space.

What if there are points in the interior of the polygon? Then dynamic programming does not apply in the same way, because each of the triangulation edges does not necessarily cut the boundary into two distinct pieces as before. Instead of only possible subregions, the number of subregions now grows exponentially. In fact, no efficient algorithm for this problem is known. More surprisingly, there is also no known proof of hardness. Minimum weight triangulation is one of the few well-studied algorithmic problems that remain in this limbo state.

Next: Limitations of Dynamic Programming Up: Dynamic Programming Previous: Longest Increasing Sequence Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE48.HTM (2 of 2) [19/1/2003 1:28:49]

Limitations of Dynamic Programming

Next: War Story: Evolution of Up: Breaking Problems Down Previous: Minimum Weight Triangulation

Limitations of Dynamic Programming Dynamic programming can be applied to any problem that observes the principle of optimality. Roughly stated, this means that partial solutions can be optimally extended with regard to the state after the partial solution instead of the partial solution itself. For example, to decide whether to extend an approximate string matching by a substitution, insertion, or deletion, we do not need to know exactly which sequence of operations was performed to date. In fact, there may be several different edit sequences that achieve a cost of C on the first p characters of pattern P and t characters of string T. Future decisions will be made based on the consequences of previous decisions, not the actual decisions themselves. Problems do not satisfy the principle of optimality if the actual operations matter, as opposed to just the cost of the operations. Consider a form of edit distance where we are not allowed to use combinations of operations in certain particular orders. Properly formulated, however, most combinatorial problems respect the principle of optimality. The biggest limitation on using dynamic programming is the number of partial solutions we must keep track of. For all of the examples we have seen, the partial solutions can be completely described by specifying the stopping places in the input. This is because the combinatorial objects being worked on (strings, numerical sequences, and polygons) all have an implicit order defined upon their elements. This order cannot be scrambled without completely changing the problem. Once the order is fixed, there are relatively few possible stopping places or states, so we get efficient algorithms. If the objects are not firmly ordered, however, we have an exponential number of possible partial solutions and are doomed to need an infeasible amount of memory. To illustrate this, consider the following dynamic programming algorithm for the traveling salesman problem, discussed in greater detail in [RND77]. Recall that solving a TSP means finding the order that visits each site exactly once, while minimizing the total distance traveled or cost paid. Let C(i,j) to be the to be the cost of the optimal tour from i to edge cost to travel directly from i to j. Define 1 that goes through each of the cities exactly once, in any order. The cost of the optimal TSP and can be computed recursively by identifying the first edge in tour is thus defined to be this sequence:

using the basis cases file:///E|/BOOK/BOOK2/NODE49.HTM (1 of 2) [19/1/2003 1:28:50]

Limitations of Dynamic Programming

This recurrence, although somewhat complicated to understand, is in fact correct. However, each partial solution is described by a vertex subset . Since there are subsets of n vertices, we require time and space to evaluate this recurrence. Whenever the input objects do not have an inherent leftright order, we are typically doomed to having an exponential-sized state space. Occasionally this is manageable - indeed, is a big improvement over enumerating all O(n!) possible TSP tours. Still, dynamic programming is most effective on well-ordered objects.

Next: War Story: Evolution of Up: Breaking Problems Down Previous: Minimum Weight Triangulation Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE49.HTM (2 of 2) [19/1/2003 1:28:50]

War Story: Evolution of the Lobster

Next: War Story: What's Past Up: Breaking Problems Down Previous: Limitations of Dynamic Programming

War Story: Evolution of the Lobster I'll confess that I'm always surprised to see graduate students stop by my office early in the morning. Something about working through the night makes that time inconvenient for them. But there they were, two future Ph.D.s working in the field of high-performance computer graphics. They studied new techniques for rendering pretty computer images. The picture they painted for me that morning was anything but pretty, however. ``You see, we want to build a program to morph one image into another,'' they explained. ``What do you mean by morph?'' I asked. ``For special effects in movies, we want to construct the intermediate stages in transforming one image into another. Suppose we want to turn you into Humphrey Bogart. For this to look realistic, we must construct a bunch of in-between frames that start out looking like you and end up looking like him.'' ``If you can realistically turn me into Bogart, you have something,'' I agreed. ``But our problem is that it isn't very realistic.'' They showed me a dismal morph between two images. ``The trouble is that we must find the right correspondence between features in the two images. It looks real bad when we get the correspondence wrong and try to morph a lip into an ear.'' ``I'll bet. So you want me to give you an algorithm for matching up lips?''

Figure: A successful alignment of two lines of pixels ``No, even simpler. We just morph each row of the initial image into the identical row of the final image. You can assume that we give you two lines of pixels, and you have to find the best possible match between the dark pixels in a row from object A to the dark pixels in the corresponding row of object B.

file:///E|/BOOK/BOOK2/NODE50.HTM (1 of 4) [19/1/2003 1:28:52]

War Story: Evolution of the Lobster

Like this,'' they said, showing me images of successful matchings like those in Figure

.

``I see,'' I said. ``You want to match big dark regions to big dark regions and small dark regions to small dark regions.'' ``Yes, but only if the matching doesn't shift them too much to the left or the right. We might prefer to merge or break up regions rather than shift them too far away, since that might mean matching a chin to an eyebrow. What is the best way to do it?'' ``One last question. Will you ever want to match two intervals to each other in such a way that they cross?'' ``No, I guess not. Crossing intervals can't match. It would be like switching your left and right eyes.'' I scratched my chin and tried to look puzzled, but I'm just not as good an actor as Bogart. I'd had a hunch about what needed to be done the instant they started talking about lines of pixels. They want to transform one array of pixels into another array, with the minimum amount of changes. That sounded like editing one string of pixels into another string, which is a classic application of dynamic programming. See Sections

and

for discussions of approximate string matching.

The fact that the intervals couldn't cross just sealed things up. It meant that whenever a stretch of dark pixels from A was mapped to a stretch from B, the problem would be split into two smaller subproblems, i.e. the pixels to the left of the match and the pixels to the right of the match. The cost of the global match would ultimately be the cost of this match plus those of matching all the pixels to the left and of matching all the pixels to the right. Constructing the optimal match on the left side is a smaller problem and hence simpler. Further, there can be only possible left subproblems, since each is completely described by the pair of one of n top pixels and one of n bottom pixels. ``Your algorithm will be based on dynamic programming,'' I pronounced. ``However, there are several possible ways to do things, depending upon whether you want to edit pixels or runs. I would probably convert each row into a list of black pixel runs, with the runs sorted by right endpoint and each run labeled with its starting position and length. You will maintain the cost of the cheapest match between the leftmost i runs and the leftmost j runs for all i and j. The possible edit operations are:

file:///E|/BOOK/BOOK2/NODE50.HTM (2 of 4) [19/1/2003 1:28:52]

War Story: Evolution of the Lobster

Figure: Morphing a lobster into a head via dynamic programming ●

●

●

Full run match: We may match run i on top to run j on the bottom for a cost that is a function of the difference in the lengths of the two runs and their positions. Merging runs: We may match a string of consecutive runs on top to a run on the bottom. The cost will be a function of the number of runs, their relative positions, and their lengths. Splitting runs: We may match a big run on top to a string of consecutive runs on the bottom. This is just the converse of the merge. Again, the cost will be a function of the number of runs, their relative positions, and their lengths.

``For each pair of runs (i,j) and all the cases that apply, we compute the cost of the edit operation and add to the (already computed and stored) edit cost to the left of the start of the edit. The cheapest of these cases is what we will take for the cost of c[i,j].'' The pair of graduate students scribbled this down, then frowned. ``So we are going to have a cost measure for matching two runs that is a function of their lengths and positions. How do we decide what the relative costs should be?''

file:///E|/BOOK/BOOK2/NODE50.HTM (3 of 4) [19/1/2003 1:28:52]

War Story: Evolution of the Lobster

``That is your business. The dynamic programming serves to optimize the matchings once you know the cost functions. It is up to your aesthetic sense to decide what penalties there should be for line length changes or offsets. My recommendation is that you implement the dynamic programming and try different values for the constants effecting the relative penalties on each of several different images. Then pick the setting that seems to do what you want.'' They looked at each other and smiled, then ran back into the lab to implement it. Using dynamic programming to do their alignments, they completed their morphing system, which is described in [HWK94]. They produced the images shown in Figure , morphing a lobster into a man. Unfortunately, they never got around to turning me into Humphrey Bogart.

Next: War Story: What's Past Up: Breaking Problems Down Previous: Limitations of Dynamic Programming Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE50.HTM (4 of 4) [19/1/2003 1:28:52]

War Story: What's Past is Prolog

Next: War Story: Text Compression Up: Breaking Problems Down Previous: War Story: Evolution of

War Story: What's Past is Prolog ``But our heuristic works very, very well in practice.'' He was simultaneously boasting and crying for help. Unification is the basic computational mechanism in logic programming languages like Prolog. A Prolog program consists of a set of rules, where each rule has a head and an associated action whenever the rule head matches or unifies with the current computation. An execution of a Prolog program starts by specifying a goal, say p(a,X,Y), where a is a constant and X and Y are variables. The system then systematically matches the head of the goal with the head of each of the rules that can be unified with the goal. Unification means binding the variables with the constants, if it is possible to match them. For the nonsense program below, p(X,Y,a) unifies with either of the first two rules, since X and Y can be bound to match the extra characters. The goal p(X,X,a) would only match the first rule, since the variable bound to the first and second positions must be the same.

p(a,a,a) := h(a);

p(b,a,a) := h(a) * h(b);

p(c,b,b) := h(b) + h(c);

p(d,b,b) := h(d) + h(b);

file:///E|/BOOK/BOOK2/NODE51.HTM (1 of 4) [19/1/2003 1:28:54]

War Story: What's Past is Prolog

Figure: Two different tries for the given set of rule heads ``In order to speed up unification, we want to preprocess the set of rule heads so that we can quickly determine which rules match a given goal. We must organize the rules in a trie data structure for fast unification.'' Tries are extremely useful data structures in working with strings, as discussed in Section . Every leaf of the trie represents one string. Each node on the path from root to leaf is labeled with exactly one character of the string, with the ith node of the path corresponding to the ith character of the string. ``I agree. A trie is a natural way to represent your rule heads. Building a trie on a set of strings of characters is straightforward - just insert the strings one after another starting from the root. So what is your problem?'' I asked. ``The efficiency of our unification algorithm depends very much on minimizing the number of edges in the trie. Since we know all the rules in advance, we have the freedom to reorder the character positions in the rules. Instead of the root node always representing the first argument in the rule, we can choose to have it represent the third argument. We would like to use this freedom to build a minimum-size trie for the set of rules.'' He showed me the example in Figure . A trie constructed according to the original string position order (1, 2, 3) uses a total of 12 edges. However, by permuting the character order to (2, 3, 1), we can obtain a trie with only 8 edges. ``Why does the speed of unification depend on minimizing the number of edges in the trie?'' ``An open goal is one with all distinct variables in its arguments, like p(X,Y,Z). Open goals match everything. In unifying an open goal against a set of clause heads, each symbol in all the clause heads will be bound to some variable. By doing a depth-first traversal of this minimum-edge trie, we minimize file:///E|/BOOK/BOOK2/NODE51.HTM (2 of 4) [19/1/2003 1:28:54]

War Story: What's Past is Prolog

the number of operations in unifying the goal with all of the clause heads,'' he explained quickly. ``Interesting...'' I started to reply before he cut me off again. ``One other constraint. For most applications we must keep the leaves of the trie ordered, so that the leaves of the underlying tree go left-to-right in the same order as the rules appear on the page.'' ``But why do you have to keep the leaves of the trie in the given order?'' I asked. ``The order of rules in Prolog programs is very, very important. If you change the order of the rules, the program returns different results.'' Then came my mission. ``We have a greedy heuristic for building good but not optimal tries, based on picking as the root the character position that minimizes the degree of the root. In other words, it picks the character position that has the smallest number of distinct characters in it. This heuristic works very, very well in practice. But we need you to prove that finding the best trie is NP-complete so our paper is, well, complete.'' I agreed to think about proving the hardness of the problem and chased him from my office. The problem did seem to involve some non-trivial combinatorial optimization to build the minimal tree, but I couldn't see how to factor the left-to-right order of the rules into a hardness proof. In fact, I couldn't think of any NP-complete problem that had such a left-right ordering constraint. After all, if a given set of rules contained a character position in common to all the rules, that character position must be probed first in any minimum-size tree. Since the rules were ordered, each node in the subtree must represent the root of a run of consecutive rules, so there were only possible nodes to choose from for this tree.... Bingo! That settled it. The next day I went back to the professor and told him. ``Look, I cannot prove that your problem is NPcomplete. But how would you feel about an efficient dynamic programming algorithm to find the best trie!'' It was a pleasure watching his frown change to a smile as the realization took hold. An efficient algorithm to compute what he needed was infinitely better than a proof saying you couldn't do it! My recurrence looked something like this. Suppose that we are given n ordered rule heads , , partitioned the rule heads into runs each with m arguments. Probing at the pth position, , where each rule in a given run had the same character value of . The rules in each run must be consecutive, so there are only

possible runs to worry about. The cost of probing at

position p is the cost of finishing the trees formed by each of the created runs, plus one edge per tree to link it to probe p:

file:///E|/BOOK/BOOK2/NODE51.HTM (3 of 4) [19/1/2003 1:28:54]

War Story: What's Past is Prolog

A graduate student immediately set to work implementing dynamic programming to compare with their heuristic. On many programs the optimal and greedy algorithms constructed the exact same trie. However, for some examples, dynamic programming gave a 20% performance improvement over greedy, i.e. 20% better than very, very well in practice. The run time spent in doing the dynamic programming was sometimes a bit larger than with greedy, but in compiler optimization you are always happy to trade off a little extra compilation time for better execution time performance of your program. Is a 20% improvement worth it? That depends upon the situation. For example, how useful would you find a 20% increase in your salary? The fact that the rules had to remain ordered was the crucial property that we exploited in the dynamic programming solution. Indeed, without it, I was able to prove that the problem was NP-complete, something we put in the paper [DRR 95] to make it complete.

Next: War Story: Text Compression Up: Breaking Problems Down Previous: War Story: Evolution of Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE51.HTM (4 of 4) [19/1/2003 1:28:54]

War Story: Text Compression for Bar Codes

Next: Divide and Conquer Up: Breaking Problems Down Previous: War Story: What's Past

War Story: Text Compression for Bar Codes Ynjiun waved his laser wand over the torn and crumpled fragments of a bar code label. The system hesitated for a few seconds, then responded with a pleasant blip sound. He smiled at me in triumph. ``Virtually indestructible.'' I was visiting the research laboratories of Symbol Technologies, of Bohemia NY, the world's leading manufacturer of bar code scanning equipment. Next time you are in the checkout line at a grocery store, check to see what type of scanning equipment they are using. Likely it will say Symbol on the housing.

Although we take bar codes for granted, there is a surprising amount of technology behind them. Bar codes exist primarily because conventional optical character recognition (OCR) systems are not sufficiently reliable for inventory operations. The bar code symbology familiar to us on each box of cereal or pack of gum encodes a ten-digit number with sufficient error correction such that it is virtually impossible to scan the wrong number, even if the can is upside-down or dented. Occasionally, the cashier won't be able to get a label to scan at all, but once you hear that blip you know it was read correctly. The ten-digit capacity of conventional bar code labels means that there is only room to store a single ID number in a label. Thus any application of supermarket bar codes must have a database mapping 1114147011 to a particular size and brand of soy sauce. The holy grail of the bar code world has long been the development of higher-capacity bar code symbologies that can store entire documents, yet still be read reliably. Largely through the efforts of Theo Pavlidis and Ynjiun Wang at Stony Brook [PSW92], Symbol Technologies was ready to introduce the first such product.

Figure: A two-dimensional bar-code label of the Gettysburg Address using PDF-417 ``PDF-417 is our new, two-dimensional bar code symbology,'' Ynjiun explained. A sample label is file:///E|/BOOK/BOOK2/NODE52.HTM (1 of 4) [19/1/2003 1:28:55]

War Story: Text Compression for Bar Codes

shown in Figure

.

``How much data can you fit in a typical one-inch square label?'' I asked him. ``It depends upon the level of error correction we use, but about 1,000 bytes. That's enough for a small text file or image,'' he said. ``Interesting. You will probably want to use some data compression technique to maximize the amount of text you can store in a label.'' See Section algorithms.

for a discussion of standard data compression

Figure: Mode switching in PDF-417 ``We do incorporate a data compaction method,'' he explained. ``We figure there are several different types of files our customers will want to use our labels for. Some files will be all in uppercase letters, while others will use mixed-case letters and numbers. We provide four different text modes in our code, each with a different subset of ASCII characters available. So long as we stay within a mode, we can describe each character using only five bits. When we have to switch modes, we issue a mode switch command first (taking an extra five bits) and then the new code.'' ``I see. So you designed the mode character sets to try to minimize the number of mode switch operations on typical text files.'' The modes are illustrated in Figure

.

``Right. We put all the digits in one mode and all the punctuation characters in another. We also included both mode shift and mode latch commands. In a mode shift, we switch into a new mode just for the next character, say to produce a punctuation mark. This way, we don't pay a cost for returning back to text mode after a period. Of course, we can also latch permanently into a different mode if we will be using a run of several characters from there.'' ``Wow!'' I said. ``With all of this mode switching going on, there must be many different ways to encode file:///E|/BOOK/BOOK2/NODE52.HTM (2 of 4) [19/1/2003 1:28:55]

War Story: Text Compression for Bar Codes

any given text as a label. How do you find the smallest such encoding.'' ``We use a greedy-type algorithm. We look a few characters ahead and then decide which mode we would be best off in. It works fairly well.'' I pressed him on this. ``How do you know it works fairly well? There might be significantly better encodings that you are simply not finding.'' ``I guess I don't know. But it's probably NP-complete to find the optimal coding.'' Ynjiun's voice trailed off. ``Isn't it?'' I started to think. Every encoding started in a given mode and consisted of a sequence of intermixed character codes and mode shift/latch operations. At any given position in the text, we could output the next character code (if it was available in our current mode) or decide to shift. As we moved from left to right through the text, our current state would be completely reflected by our current character position and current mode state. For a given position/mode pair, we would have been interested in the cheapest way of getting there, over all possible encodings getting to this point.... My eyes lit up so bright they cast shadows on the walls. ``The optimal encoding for any given text in PDF-417 can be found using dynamic programming. For each possible mode and each character position , we will maintain the cheapest encoding found of the string to the left of i ending in mode m. From each mode/position, we can either match, shift, or latch, so there are only few possible operations to consider. Each of the 4n cells can be filled in constant time, so it takes time linear in the length of the string to find the optimal encoding.'' Ynjiun was skeptical, but he encouraged us to implement an optimal encoder. A few complications arose due to weirdnesses of PDF-417 mode switching, but my student Yaw-Ling Lin rose to the challenge. Symbol compared our encoder to theirs on 13,000 labels and concluded that dynamic programming lead to an 8% tighter encoding on average. This was significant, because no one wants to waste 8% of their potential storage capacity, particularly in an environment where the capacity is only a few hundred bytes. For certain applications, this 8% margin permitted one bar code label to suffice where previously two had been required. Of course, an 8% average improvement meant that it did much better than that on certain labels. While our encoder took slightly longer to run than the greedy encoder, this was not significant, since the bottleneck would be the time needed to print the label, anyway. Our observed impact of replacing a heuristic solution with the global optimum is probably typical of most applications. Unless you really botch your heuristic, you are probably going to get a decent solution. But replacing it with an optimal result usually gives a small but non-trivial improvement, which can have pleasing consequences for your application.

file:///E|/BOOK/BOOK2/NODE52.HTM (3 of 4) [19/1/2003 1:28:55]

War Story: Text Compression for Bar Codes

Next: Divide and Conquer Up: Breaking Problems Down Previous: War Story: What's Past Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE52.HTM (4 of 4) [19/1/2003 1:28:55]

Divide and Conquer

Next: Fast Exponentiation Up: Breaking Problems Down Previous: War Story: Text Compression

Divide and Conquer Divide and conquer was a successful military strategy long before it became an algorithm design paradigm. Generals observed that it was easier to defeat one army of 50,000 men, followed by another army of 50,000 men than it was to beat a single 100,000 man army. Thus the wise general would attack so as to divide the enemy army into two forces and then mop up one after the other. To use divide and conquer as an algorithm design technique, we must divide the problem into two smaller subproblems, solve each of them recursively, and then meld the two partial solutions into one solution to the full problem. Whenever the merging takes less time than solving the two subproblems, we get an efficient algorithm. Mergesort, discussed in Section , is the classic example of a divide-andconquer algorithm. It takes only linear time to merge two sorted lists of n/2 elements each of which was obtained in time. Divide and conquer is a design technique with many important algorithms to its credit, including mergesort, the fast Fourier transform, and Strassen's matrix multiplication algorithm. However, with the exception of binary search, I find it to be a difficult technique to apply in practice. Therefore, the examples below will illustrate binary search and its variants, which yield simple algorithms to solve a surprisingly wide range of problems.

● ● ●

Fast Exponentiation Binary Search Square and Other Roots

Algorithms

file:///E|/BOOK/BOOK2/NODE53.HTM (1 of 2) [19/1/2003 1:28:56]

Divide and Conquer

Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE53.HTM (2 of 2) [19/1/2003 1:28:56]

Fast Exponentiation

Next: Binary Search Up: Divide and Conquer Previous: Divide and Conquer

Fast Exponentiation Suppose that we need to compute the value of

for some reasonably large n. Such problems occur in

primality testing for cryptography, as discussed in Section

.

The simplest algorithm performs n-1 multiplications, by computing . However, we can do better by observing that . If n is even, then . If n is odd, then . In either case, we have halved the size of our exponent at the cost of at most two multiplications, so multiplications suffice to compute the final value.

function power(a,n)

if (n = 0) return(1)

x = power(

)

if (n is even) then return(

else return(

)

)

This simple algorithm illustrates an important principle of divide and conquer. It always pays to divide a job as evenly as possible. This principle applies to real life as well. When n is not a power of two, the problem cannot always be divided perfectly evenly, but a difference of one element between the two sides cannot cause any serious imbalance.

file:///E|/BOOK/BOOK2/NODE54.HTM (1 of 2) [19/1/2003 1:28:57]

Fast Exponentiation

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE54.HTM (2 of 2) [19/1/2003 1:28:57]

Binary Search

Next: Square and Other Roots Up: Divide and Conquer Previous: Fast Exponentiation

Binary Search Binary search is a fast algorithm for searching in a sorted array S of keys. To search for key q, we compare q to the middle key S[n/2]. If q appears before S[n/2], it must reside in the top half of our set; if not, it must reside in the bottom half of our set. By recursively repeating this process on the correct half, we find the key in a total of comparisons, a big win over the n/2 we expect with sequential search. This much you probably know. What is important is to have a sense for just how fast binary search is. Twenty questions is a popular children's game, where one player selects a word, and the other repeatedly asks true/false questions in an attempt to identify the word. If the word remains unidentified after 20 questions, the first party wins; otherwise, the second player takes the honors. In fact, the second player always has a winning strategy, based on binary search. Given a printed dictionary, the player opens it in the middle, selects a word (say ``move''), and asks whether the unknown word is before ``move'' in alphabetical order. Since standard dictionaries contain 50,000 to 200,000 words, we can be certain that the process will always terminate within twenty questions. Other interesting algorithms follow from simple variants of binary search. For example, suppose we have an array A consisting of a run of 0's, followed by an unbounded run of 1's, and would like to identify the exact point of transition between them. Binary search on the array would provide the transition point in tests, if we had a bound n on the number of elements in the array. In the absence of such a bound, we can test repeatedly at larger intervals (A[1], A[2], A[4], A[8], A[16], ) until we find a first non-zero value. Now we have a window containing the target and can proceed with binary search. This onesided binary search finds the transition point p using at most comparisons, regardless of how large the array actally is. One-sided binary search is most useful whenever we are looking for a key that probably lies close to our current position.

Next: Square and Other Roots Up: Divide and Conquer Previous: Fast Exponentiation Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE55.HTM [19/1/2003 1:28:58]

Square and Other Roots

Next: Exercises Up: Divide and Conquer Previous: Binary Search

Square and Other Roots The square root of n is the number r such that . Square root computations are performed inside every pocket calculator, but it is instructive to develop an efficient algorithm to compute square roots. First, observe that the square root of

must be at least 1 and at most n. Let l=1 and r=n. Consider the

midpoint of this interval, m=(l+r)/2. How does

compare to n? If

, then the square root must be

greater than m, so the algorithm repeats with l=m. If , then the square root must be less than m, so the algorithm repeats with r=m. Either way, we have halved the interval with only one comparison. rounds we will have identified the square root to within . Therefore, after only This bisection method, as it is called in numerical analysis, can also be applied to the more general problem of finding the roots of an equation. We say that x is a root of the function f if f(x)=0. Suppose that we start with values l and r such that f(l) > 0 and f(r) < 0. If f is a continuous function, there must be a root between l and r. Depending upon the sign of f(m), where m=(l+r)/2, we can cut this window containing the root in half with each test and stop when our estimate becomes sufficiently accurate. Root-finding algorithms that converge faster than binary search are known for both of these problems. Instead of always testing the midpoint of the interval, these algorithms interpolate to find a test point closer to the actual root. Still, binary search is simple, robust, and works as well as possible without additional information on the nature of the function to be computed.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE56.HTM [19/1/2003 1:28:59]

Exercises

Next: Implementation Challenges Up: Breaking Problems Down Previous: Square and Other Roots

Exercises 1. Consider the problem of storing n books on shelves in a library. The order of the books is fixed by the cataloging system and so cannot be rearranged. Therefore, we can speak of a book , where , that has a thickness and height . The length of each bookshelf at this library is L. Suppose all the books have the same height h (i.e. for all i, j) and the shelves are all separated by a distance of greater than h, so any book fits on any shelf. The greedy algorithm would fill the first shelf with as many books as we can until we get the smallest i such that does not fit, and then repeat with subsequent shelves. Show that the greedy algorithm always finds the optimal shelf placement, and analyze its time complexity. 2. (*) This is a generalization of the previous problem. Now consider the case where the height of the books is not constant, but we have the freedom to adjust the height of each shelf to that of the tallest book on the shelf. Thus the cost of a particular layout is the sum of the heights of the largest book on each shelf. Give an example to show that the greedy algorithm of stuffing each shelf as full as possible does not always give the minimum overall height. ❍ Give an algorithm for this problem, and analyze its time complexity. Hint: use dynamic programming. 3. (*) Consider a city whose streets are defined by an grid. We are interested in walking from the upper left-hand corner of the grid to the lower right-hand corner. ❍

Unfortunately, the city has bad neighborhoods, which are defined as intersections we do not want to walk in. We are given an matrix BAD, where BAD[i,j] = ``yes'' if and only if the intersection between streets i and j is somewhere we want to avoid. (a) Give an example of the contents of BAD such that there is no path across the grid avoiding bad neighborhoods. (b) Give an O( X Y ) algorithm to find a path across the grid that avoids bad neighborhoods.

file:///E|/BOOK/BOOK2/NODE57.HTM (1 of 4) [19/1/2003 1:29:01]

Exercises

(c) Give an O( X Y ) algorithm to find the shortest path across the grid that avoids bad neighborhoods. You may assume that all blocks are of equal length. For partial credit, give an algorithm. 4. (*) Consider the same situation as the previous problem. We have a city whose streets are defined grid. We are interested in walking from the upper left-hand corner of the grid to the by an lower right-hand corner. We are given an matrix BAD, where BAD[i,j] = ``yes'' if and only if the intersection between streets i and j is somewhere we want to avoid. If there were no bad neighborhoods to contend with, the shortest path across the grid would have length (X-1) + (Y-1) blocks, and indeed there would be many such paths across the grid. Each path would consist of only rightward and downward moves. Give an algorithm that takes the array BAD and returns the number of safe paths of length X+Y-2. For full credit, your algorithm must run in O( X Y ). 5. (*) Given an array of n real numbers, consider the problem of finding the maximum sum in any contiguous subvector of the input. For example, in the array

the maximum is achieved by summing the third through seventh elements, where 59+26+(53)+58+97 = 187. When all numbers are positive, the entire array is the answer, while when all numbers are negative, the empty array maximizes the total at 0. Give a simple, clear, and correct -time algorithm to find the maximum contiguous subvector. ❍ Now give a -time dynamic programming algorithm for this problem. To get partial credit, you may instead give a correct divide-and-conquer algorithm. 6. In the United States, coins are minted with denominations of 1, 5, 10, 25, and 50 cents. Now consider a country whose coins are minted with denominations of units. They seek an algorithm that will enable them to make change of n units using the minimum number of coins. ❍

(a) The greedy algorithm for making change repeatedly uses the biggest coin smaller than the amount to be changed until it is zero. Show that the greedy algorithm does not always give the minimum number of coins in a country whose denominations are . (b) Give an efficient algorithm that correctly determines the minimum number of coins needed to make change of n units using denominations . Analyze its running time. 7. (*) In the United States, coins are minted with denominations of 1, 5, 10, 25, and 50 cents. Now units. They want to consider a country whose coins are minted with denominations of count how many distinct ways C(n) there are to make change of n units. For example, in a country

file:///E|/BOOK/BOOK2/NODE57.HTM (2 of 4) [19/1/2003 1:29:01]

Exercises

whose denominations are , C(5) = 1, C(6) to C(9)=2, C(10)=3, and C(12)=4. ? 1. How many ways are there to make change of 20 units from 2. Give an efficient algorithm to compute C(n), and analyze its complexity. (Hint: think in terms of computing C(n,d), the number of ways to make change of n units with highest denomination d. Be careful to avoid overcounting.) of characters from an alphabet on 8. (**) Consider the problem of examining a string k symbols, and a multiplication table over this alphabet, and deciding whether or not it is possible to parenthesize x in such a way that the value of the resulting expression is a, where a belongs to the alphabet. The multiplication table is neither commutative or associative, so the order of multiplication matters. a b c a a c c b a a b c c c c For example, consider the following multiplication table and the string bbbba. Parenthesizing it (b(bb))(ba) gives a, but ((((bb)b)b)a) gives c. Give an algorithm, with time polynomial in n and k, to decide whether such a parenthesization exists for a given string, multiplication table, and goal element. 9. (*) Consider the following data compression technique. We have a table of m text strings, each of length at most k. We want to encode a data string D of length n using as few text strings as possible. For example, if our table contains (a,ba,abab,b) and the data string is bababbaababa, the best way to encode it is (b,abab,ba,abab,a) - a total of five code words. Give an O(nmk) algorithm to find the length of the best encoding. You may assume that the string has an encoding in terms of the table. 10. A company database consists of 10,000 sorted names, 40% of whom are known as good customers and who together account for 60% of the accesses to the data base. There are two data structure options to consider for representing the database: ❍ ❍

Put all the names in a single array and use binary search. Put the good customers in one array and the rest of them in a second array. Only if we do not find the query name on a binary search of the first array do we do a binary search of the second array.

Demonstrate which option gives better expected performance. Does this change if linear search on an unsorted array is used instead of binary search for both options? 11. Suppose you are given an array A of n sorted numbers that has been circularly shifted k positions to the right. For example, is a sorted array that has been circularly shifted k=2

file:///E|/BOOK/BOOK2/NODE57.HTM (3 of 4) [19/1/2003 1:29:01]

Exercises

positions, while

has been shifted k=4 positions.

Suppose you know what k is. Give an O(1) algorithm to find the largest number in A. ❍ Suppose you do not know what k is. Give an algorithm to find the largest number in A. For partial credit, you may give an O(n) algorithm. 12. (*) Suppose that you are given a sorted sequence of distinct integers . Give an algorithm to determine whether there exists an index i such at . For example, in , . In , there is no such i. ❍

Next: Implementation Challenges Up: Breaking Problems Down Previous: Square and Other Roots Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE57.HTM (4 of 4) [19/1/2003 1:29:01]

The Friendship Graph

Next: Data Structures for Graphs Up: Graph Algorithms Previous: Graph Algorithms

The Friendship Graph

Figure: A portion of the friendship graph To demonstrate the importance of proper modeling, let us consider a graph where the vertices are people, and there is an edge between two people if and only if they are friends. This graph is well-defined on any set of people, be they the people in your neighborhood, at your school or place of business, or even the entire world. There are many interesting aspects of people that are best understood as properties of this friendship graph. We use this opportunity to define important graph theory terminology. ``Talking the talk'' proves to be an important part of ``walking the walk''. ●

If I am your friend, does that mean you are my friend?

●

A graph is undirected if edge (x,y) always implies (y,x). Otherwise, the graph is said to be directed. The ``heard-of'' graph is directed, since many famous people who I have heard of have never heard of me! The ``had-sex-with'' graph is presumably undirected, since the critical operation always requires a partner. I'd like to think that the ``friendship'' graph is always an undirected graph. Am I my own friend? An edge of the form (x,x) is said to be a loop. If x was y's friend several times over, we can model this relationship using multiedges, multiple edges between the same pair of vertices. A graph is said to be simple if it contains no loops and no multiple edges. Simple graphs really are often simpler to work with in practice. Therefore, we might be better off if no one was their own friend.

●

How close a friend are you? A graph is said to be weighted if each edge has an associated numerical attribute. We could model

file:///E|/BOOK/BOOK2/NODE60.HTM (1 of 3) [19/1/2003 1:29:02]

The Friendship Graph

the strength of a friendship by associating each edge with an appropriate number, say from 0 (enemies) to 10 (blood brothers). The edges of a road network graph might be weighted with their length, drive-time, or speed limit, depending upon the application. A graph is said to be unweighted if all edges are assumed to be of equal weight.

Figure: Mel Brooks is my father's sister's husband's cousin ●

Am I linked by some chain of friends to a star? A path is a sequence of edges connecting two vertices. Since Mel Brooks is my father's sister's husband's cousin, there is a path in the friendship graph between me and him, shown in Figure

●

. This is true even though the two of us have never met. How close is my link to that star?

●

If I were trying to impress you with how tight I am with Mel Brooks, I would be much better off saying that my Uncle Lenny grew up with him than to go into the details of how connected I am to Uncle Lenny. Through Uncle Lenny, I have a path of length 2 to Cousin Mel, while the path is of length 4 by blood and marriage. I could make the path even longer by linking in people who know both me and my father, or are friends of Aunt Eve and Uncle Lenny. This multiplicity of paths hints at why finding the shortest path between two nodes is important and instructive, even in non-transportation applications. Is there a path of friends between every two people in the world? The ``six degrees of separation'' theory argues that there is always a short path linking every two people in the world. We say that a graph is connected if there is a path between any two vertices. A directed graph is strongly connected if there is always a directed path between any two vertices.

If a graph is not connected, we call each connected piece a connected component. If we envision tribes in remote parts of the world that have yet not been encountered, each such tribe would form a connected component in the friendship graph. A remote hermit, or extremely unpleasant fellow ●

(see Figure ) would represent a connected component of one vertex, or an isolated vertex. Who has the most friends? The fewest friends? The degree of a vertex is the number of edges adjacent to it. The most popular person defines the vertex of highest degree in the friendship graph. Remote hermits will be associated with degreezero vertices. In dense graphs, most vertices have high degree, as opposed to sparse graphs with relatively few edges. In regular graphs, each vertex has exactly the same degree. A regular

file:///E|/BOOK/BOOK2/NODE60.HTM (2 of 3) [19/1/2003 1:29:02]

The Friendship Graph

●

friendship graph would truly be the ultimate in social-ism. What is the largest clique?

●

A social clique is a group of mutual friends who all hang around together. A graph-theoretic clique is a complete subgraph, where each vertex pair has an edge between them. Cliques are the densest possible subgraphs. Within the friendship graph, we would expect to see large cliques corresponding to workplaces, neighborhoods, religious organizations, and schools. How long will it take for my gossip to get back to me? A cycle is a path where the last vertex is adjacent to the first. A cycle in which no vertex repeats (such as 1-2-3-1 verus 1-2-3-2-1) is said to be simple. The shortest cycle in the graph defines the graph's girth, while a simple cycle that passes through every vertex once is said to be a Hamiltonian cycle. An undirected graph with no cycles is said to be a tree if it is connected; otherwise it is a forest. A directed graph with no directed cycles is said to be a DAG, or directed acyclic graph.

Next: Data Structures for Graphs Up: Graph Algorithms Previous: Graph Algorithms Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE60.HTM (3 of 3) [19/1/2003 1:29:02]

Data Structures for Graphs

Next: War Story: Getting the Up: Graph Algorithms Previous: The Friendship Graph

Data Structures for Graphs Selecting the right data structure to represent graphs can have an enormous impact on the performance of an algorithm. Your two basic choices are adjacency matrices and adjacency lists, illustrated in Figure . An adjacency matrix is an matrix M where (typically) M[i,j] = 1 if there is an edge from vertex i to vertex j and M[i,j]=0 if there is not. Adjacency matrices are the simplest way to represent graphs. However, they doom you to using space no matter how many edges are in the graph. For large 1,000,000, and work up from there. Although there is graphs, this will kill you. Remember that some potential for saving space by packing multiple bits per word or simulating a triangular matrix for undirected graphs, these cost some of the simplicity that makes adjacency matrices so appealing.

Figure: The adjacency matrix and adjacency list of a given graph Beyond simplicity, there are certain algorithmic operations that prove faster on adjacency matrices than adjacency lists. In particular, it takes time to test whether edge (i,j) is in a graph represented by an adjacency matrix. All we must do is read the appropriate bit. An adjacency list consists of an n-element array of pointers, where the ith element points to a linked list of the edges incident on vertex i. To test whether edge (i,j) is in the graph, we search the ith list for j. This takes , where is the degree of the ith vertex. For a complete or almost complete graph, , so testing the existence of an edge can be very expensive relative to adjacency matrices. However, can be much less than n when the graph is sparse. Most of the graphs that one encounters in real life tend to be sparse. Recall the friendship graph as an example. Further, a surprising number of the most efficient graph algorithms can be and have been designed to avoid such edge-existence queries. The key is processing the edges in a systematic order like breadth-first or depth-first search. For most applications, adjacency lists are the right way to go. The main drawback is the complexity of dealing with linked list structures. Things can be made arbitrarily hairy by adding extra pointers for special purposes. For example, the two versions of each edge in an undirected graph, (i,j) and (j,i), can be file:///E|/BOOK/BOOK2/NODE61.HTM (1 of 2) [19/1/2003 1:29:04]

Data Structures for Graphs

linked together by a pointer to facilitate deletions. Also, depending upon the operations you will perform on each list, you may or may not want it to be doubly linked, so that you can move backwards as easily as you move forwards. It is a good idea to use a well-designed graph data type as a model for building your own, or even better as the foundation for your application. We recommend LEDA (see Section ) as the best-designed general-purpose graph data structure currently available. It may be more powerful (and hence somewhat slower/larger) than what you need, but it does so many things right that you are likely to lose most of the potential do-it-yourself benefits through clumsiness. In summary, we have the following tradeoffs between adjacency lists and matrices: Comparison Faster to test if (x, y) is in graph?

Winner adjacency matrices

Faster to find the degree of a vertex? Less memory on small graphs?

adjacency lists adjacency lists (m+n) vs.

Less memory on big graphs?

adjacency matrices (a small win)

Edge insertion or deletion?

adjacency matrices O(1) vs. O(d)

Faster to traverse the graph?

adjacency lists

Better for most problems?

vs. adjacency lists

Next: War Story: Getting the Up: Graph Algorithms Previous: The Friendship Graph Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE61.HTM (2 of 2) [19/1/2003 1:29:04]

War Story: Getting the Graph

Next: Traversing a Graph Up: Graph Algorithms Previous: Data Structures for Graphs

War Story: Getting the Graph ``It takes five minutes just to read the data. We will never have time to make it do something interesting.'' The young graduate student was bright and eager, but green to the power of data structures. She would soon come to appreciate the power.

Figure: The dual graph (dashed lines) of a triangulation As described in a previous war story (see Section ), we were experimenting with algorithms for extracting triangular strips for the fast rendering of triangulated surfaces. The task of finding a small number of strips that cover each triangle in a mesh could be modeled as a graph problem, where the graph has a vertex for every triangle of the mesh, and there is an edge between every pair of vertices representing adjacent triangles. This dual graph representation of the planar subdivision representing the triangulation (see Figure into triangle strips.

) captures all the information about the triangulation needed to partition it

The first step in crafting a program that constructs a good set of strips was to build the dual graph of the triangulation. This I sent the student off to do. A few days later, she came back and announced that it took over five CPU minutes just to construct this dual graph of an object with a few thousand triangles. ``Nonsense!'' I proclaimed. ``You must be doing something very wasteful in building the graph. What format is the data in?''

file:///E|/BOOK/BOOK2/NODE62.HTM (1 of 3) [19/1/2003 1:29:05]

War Story: Getting the Graph

``Well, it starts out with a list of the 3D-coordinates of the vertices used in the model and then follows with a list of triangles. Each triangle is described by a list of three indices into the vertex coordinates. Here is a small example:'' VERTICES 4 0.000000 240.000000 0.000000 204.000000 240.000000 0.000000 204.000000 0.000000 0.000000 0.000000 0.000000 0.000000 TRIANGLES 2 0 1 3 1 2 3 ``I see. So the first triangle must use all but the third point, since all the indices start from zero. The two triangles must share an edge formed by points 1 and 3.'' ``Yeah, that's right,'' she confirmed. ``OK. Now tell me how you built your dual graph from this file.'' ``Well, I can pretty much ignore the vertex information, once I know how many vertices there will be. The geometric position of the points doesn't affect the structure of the graph. My dual graph is going to have as many vertices as the number of triangles. I set up an adjacency list data structure with that many vertices. As I read in each triangle, I compare it to each of the others to check whether it has two numbers in common. Whenever it does, I add an edge from the new triangle to this one.'' I started to sputter. ``But that's your problem right there! You are comparing each triangle against every other triangle, so that constructing the dual graph will be quadratic in the number of triangles. Reading in the input graph should take linear time!'' ``I'm not comparing every triangle against every other triangle. On average, it only tests against half or a third of the triangles.'' ``Swell. But that still leaves us with an

algorithm. That is much too slow.''

She stood her ground. ``Well, don't just complain. Help me fix it!'' Fair enough. I started to think. We needed some quick method to screen away most of the triangles that would not be adjacent to the new triangle (i,j,k). What we really needed was just a list of all the triangles that go through each of the points i, j, and k. Since each triangle goes through three points, the average point is incident on three triangles, so this would mean comparing each new triangle against fewer than ten others, instead of most of them. file:///E|/BOOK/BOOK2/NODE62.HTM (2 of 3) [19/1/2003 1:29:05]

War Story: Getting the Graph

``We are going to need a data structure consisting of an array with one element for every vertex in the original data set. This element is going to be a list of all the triangles that pass through that vertex. When we read in a new triangle, we will look up the three relevant lists in the array and compare each of these against the new triangle. Actually, only two of the three lists are needed, since any adjacent triangle will share two points in common. For anything sharing two vertices, we will add an adjacency to our graph. Finally, we will add our new triangle to each of the three affected lists, so they will be updated for the next triangle read.'' She thought about this for a while and smiled. ``Got it, Chief. I'll let you know what happens.'' The next day she reported that the graph could be built in seconds, even for much larger models. From here, she went on to build a successful program for extracting triangle strips, as reported in Section

.

The take-home lesson here is that even elementary problems like initializing data structures can prove to be bottlenecks in algorithm development. Indeed, most programs working with large amounts of data have to run in linear or almost linear time. With such tight performance demands, there is no room to be sloppy. Once you focus on the need for linear-time performance, an appropriate algorithm or heuristic can usually be found to do the job.

Next: Traversing a Graph Up: Graph Algorithms Previous: Data Structures for Graphs Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE62.HTM (3 of 3) [19/1/2003 1:29:05]

Traversing a Graph

Next: Breadth-First Search Up: Graph Algorithms Previous: War Story: Getting the

Traversing a Graph Perhaps the most fundamental graph problem is to traverse every edge and vertex in a graph in a systematic way. Indeed, most of the basic algorithms you will need for bookkeeping operations on graphs will be applications of graph traversal. These include: ● ● ● ● ●

Printing or validating the contents of each edge and/or vertex. Copying a graph, or converting between alternate representations. Counting the number of edges and/or vertices. Identifying the connected components of the graph. Finding paths between two vertices, or cycles if they exist.

Since any maze can be represented by a graph, where each junction is a vertex and each hallway an edge, any traversal algorithm must be powerful enough to get us out of an arbitrary maze. For efficiency, we must make sure we don't get lost in the maze and visit the same place repeatedly. By being careful, we can arrange to visit each edge exactly twice. For correctness, we must do the traversal in a systematic way to ensure that we don't miss anything. To guarantee that we get out of the maze, we must make sure our search takes us through every edge and vertex in the graph. The key idea behind graph traversal is to mark each vertex when we first visit it and keep track of what we have not yet completely explored. Although bread crumbs or unraveled threads are used to mark visited places in fairy-tale mazes, we will rely on Boolean flags or enumerated types. Each vertex will always be in one of the following three states: ● ●

●

undiscovered - the vertex in its initial, virgin state. discovered - the vertex after we have encountered it, but before we have checked out all its incident edges. completely-explored - the vertex after we have visited all its incident edges.

Obviously, a vertex cannot be completely-explored before we discover it, so over the course of the traversal the state of each vertex progresses from undiscovered to discovered to completely-explored. We must also maintain a structure containing all the vertices that we have discovered but not yet file:///E|/BOOK/BOOK2/NODE63.HTM (1 of 2) [19/1/2003 1:29:05]

Traversing a Graph

completely explored. Initially, only a single start vertex is considered to have been discovered. To completely explore a vertex, we must evaluate each edge going out of it. If an edge goes to an undiscovered vertex, we mark it discovered and add it to the list of work to do. If an edge goes to a completely-explored vertex, we will ignore it, since further contemplation will tell us nothing new about the graph. We can also ignore any edge going to a discovered but not completely-explored vertex, since the destination must already reside on the list of vertices to completely explore. Regardless of which order we use to fetch the next vertex to explore, each undirected edge will be considered exactly twice, once when each of its endpoints is explored. Directed edges will be consider only once, when exploring the source vertex. Every edge and vertex in the connected component must eventually be visited. Why? Suppose the traversal didn't visit everything, meaning that there exists a vertex u that remains unvisited whose neighbor v was visited. This neighbor v will eventually be explored, and we will certainly visit u when we do so. Thus we must find everything that is there to be found. The order in which we explore the vertices depends upon the container data structure used to store the discovered but not completely-explored vertices. There are two important possibilities: ●

●

● ●

Queue - by storing the vertices in a first in, first out (FIFO) queue, we explore the oldest unexplored vertices first. Thus our explorations radiate out slowly from the starting vertex, defining a so-called breadth-first search. Stack - by storing the vertices in a last in, first out (LIFO) stack, we explore the vertices by lurching along a path, visiting a new neighbor if one is available, and backing up only when we are surrounded by previously discovered vertices. Thus our explorations quickly wander away from our starting point, defining a so-called depth-first search.

Breadth-First Search Depth-First Search

Next: Breadth-First Search Up: Graph Algorithms Previous: War Story: Getting the Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE63.HTM (2 of 2) [19/1/2003 1:29:05]

Breadth-First Search

Next: Depth-First Search Up: Traversing a Graph Previous: Traversing a Graph

Breadth-First Search

Figure: An undirected graph and its breadth-first search tree The basic breadth-first search algorithm is given below. At some point during the traversal, every node in the graph changes state from undiscovered to discovered. In a breadth-first search of an undirected graph, we assign a direction to each edge, from the discoverer u to the discovered v. We thus denote u to be the parent p[v]. Since each node has exactly one parent, , defines a shortest path except for the root, this defines a tree on the vertices of the graph. This tree, illustrated in Figure from the root to every other node in the tree. This property makes breadth-first search very useful in shortest path problems.

BFS(G,s)

do

for each vertex

state[u] = ``undiscovered''

p[u] = nil, i.e. no parent is in the BFS tree

state[] = ``discovered''

p[] = nil

while

do

u =

dequeue[Q]

process vertex u as desired

file:///E|/BOOK/BOOK2/NODE64.HTM (1 of 2) [19/1/2003 1:29:06]

Breadth-First Search

for each

do

process edge (u,v) as desired

if state[v] = ``undiscovered'' then

state[v] = ``discovered''

p[v] = u

enqueue[Q,v]

state[u] = ``completely-explored''

The graph edges that do not appear in the breadth-first search tree also have special properties. For undirected graphs, nontree edges can point only to vertices on the same level as the parent vertex or to vertices on the level directly below the parent. These properties follow easily from the fact that each path in the tree must be the shortest path in the graph. For a directed graph, a back-pointing edge

can exist whenever v lies closer to the root than u does.

The breadth-first search algorithm above includes places to optionally process each vertex and edge, say to copy them, print them, or count them. Each vertex and directed edge is encountered exactly once, and each undirected edge is encountered exactly twice.

Next: Depth-First Search Up: Traversing a Graph Previous: Traversing a Graph Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE64.HTM (2 of 2) [19/1/2003 1:29:06]

Depth-First Search

Next: Applications of Graph Traversal Up: Traversing a Graph Previous: Breadth-First Search

Depth-First Search

Figure: An undirected graph and its depth-first search tree Depth-first search turns out to be, in general, even more useful than breadth-first search. The reason is that a depth-first search of a graph organizes the edges of the graph in a very precise way, which is quite different from breadth-first search. As with BFS, we assign a direction to each edge when we discover it, as shown in Figure

.

Figure: Edge possibilities for search trees , only two of them Although there are four conceivable classes of edges resulting from such labelings, as shown in Figure can occur with undirected graphs. In a DFS of an undirected graph, every edge is either in the tree or goes directly back to an ancestor. Why? Suppose we encountered a forward edge (x,y) directed toward a decendant vertex. In this case, we would have discovered (x,y) when exploring y, making it a back edge. Suppose we encounter a cross edge (x,y), linking two unrelated vertices. Again, we would have discovered this edge when we explored y, making it a tree edge. For directed graphs, depth-first search labelings can take on a wider range of possibilities. Depth-first search has a neat recursive implementation, which eliminates the need to explicitly use a stack:

DFS(G,u)

state[u] = ``discovered''

process vertex u if desired

file:///E|/BOOK/BOOK2/NODE65.HTM (1 of 3) [19/1/2003 1:29:07]

Depth-First Search

for each

do

process edge (u,v) if desired

if state[v] = ``undiscovered'' then

p[v] = u

DFS(G,v)

state[u] = ``completely-explored''

As with BFS, this implementation of the depth-first search algorithm includes places to optionally process each vertex and edge, say to copy them, print them, or count them. Both algorithms will traverse all edges in the same connected component as the starting point. Since we need to start with a vertex in each component in order to traverse a disconnected graph, we must start from any vertex remaining undiscovered after a component search. With the proper initialization, this completes the traversal algorithm:

DFS-graph(G)

for each vertex

do

state[u] = ``undiscovered''

for each vertex

do

if state[u] = ``undiscovered'' then

initialize new component, if desired

DFS[G,u]

file:///E|/BOOK/BOOK2/NODE65.HTM (2 of 3) [19/1/2003 1:29:07]

Depth-First Search

Next: Applications of Graph Traversal Up: Traversing a Graph Previous: Breadth-First Search Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE65.HTM (3 of 3) [19/1/2003 1:29:07]

Applications of Graph Traversal

Next: Connected Components Up: Graph Algorithms Previous: Depth-First Search

Applications of Graph Traversal Most elementary graph algorithms involve making one or two traversals of the graph, while we update our knowledge of the graph as we visit each edge and vertex. Properly implemented using adjacency lists, any such algorithm is destined to be very fast. Both BFS and DFS run in O(n+m) on both directed and undirected graphs where, as usual, n is the number of vertices and m the number of edges in the graph. This is optimal, since it is as fast as one can hope to read the graph. The trick is seeing when traversal approaches are destined to work. We present several examples below.

● ● ● ● ●

Connected Components Tree and Cycle Detection Two-Coloring Graphs Topological Sorting Articulation Vertices

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE66.HTM [19/1/2003 1:29:08]

Connected Components

Next: Tree and Cycle Detection Up: Applications of Graph Traversal Previous: Applications of Graph Traversal

Connected Components Either breadth-first or depth-first search can be used to identify the connected components of an undirected graph and label each vertex with the identifier of its components. In particular, we can modify the DFS-graph algorithm to increment a counter for the current component number and label each vertex accordingly as it is discovered in DFS. For directed graphs, there are two distinct notions of connectivity, leading to algorithms for finding both weakly connected and strongly connected components. Both of these can be found in O(n+m) time, as discussed in Section

.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE67.HTM [19/1/2003 1:29:08]

Tree and Cycle Detection

Next: Two-Coloring Graphs Up: Applications of Graph Traversal Previous: Connected Components

Tree and Cycle Detection Trees are connected, undirected graphs that do not contain cycles. They are perhaps the simplest interesting class of graphs. Testing whether a graph is a tree is straightforward using depth-first search. During search, every edge will be labeled either a tree edge or a back edge, so the graph is a tree if and only if there are no back edges. Since m=n-1 for any tree, this algorithm can be said to run in time linear in the number of vertices. If the graph is not a tree, it must contain a cycle. Such a cycle can be identified as soon as the first back edge (u,v) is detected. If (u,v) is a back edge, then there must be a path in the tree from v to u. Coupled with edge (u,v), this defines a cycle.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE68.HTM [19/1/2003 1:29:09]

Two-Coloring Graphs

Next: Topological Sorting Up: Applications of Graph Traversal Previous: Tree and Cycle Detection

Two-Coloring Graphs In vertex coloring, we seek to assign a color to each vertex of a graph G such that no edge links two vertices of the same color. We can avoid all conflicts by assigning each vertex its own color. However, the goal is to use as few colors as possible. Vertex coloring problems arise often in scheduling applications, such as register allocation in compilers. See Section coloring algorithms and applications.

for a full treatment of vertex

A graph is bipartite if it can be colored without conflicts while using only two colors. Bipartite graphs are important because they arise often in practice and have more structure than arbitrary graphs. For example, consider the ``had-sex-with'' graph in a heterosexual world. Men have sex only with women, and vice versa. Thus gender defines a legal two-coloring. Irrespective of the accuracy of the model, it should be clear that bipartite graphs are simpler to work with than general graphs. But how can we find an appropriate two-coloring of a graph, thus separating the men from the women? Suppose we assume that the starting vertex is male. All vertices adjacent to this man must be female, assuming the graph is indeed bipartite. We can augment either breadth-first or depth-first search so that whenever we discover a new vertex, we color it the opposite of its parent. For each non-discovery edge, we check whether it links two vertices of the same color. Such a conflict means that the graph cannot be two-colored. However, we will have constructed a proper two-coloring whenever we terminate without conflict. We can assign the first vertex in any connected component to be whatever color/sex we wish. Although we can separate the men from the women, we can't tell them apart just using the graph.

Next: Topological Sorting Up: Applications of Graph Traversal Previous: Tree and Cycle Detection Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE69.HTM [19/1/2003 1:29:09]

Topological Sorting

Next: Articulation Vertices Up: Applications of Graph Traversal Previous: Two-Coloring Graphs

Topological Sorting

Figure: Directed acyclic and cyclic graphs A directed, acyclic graph, or DAG, is a directed graph with no directed cycles. Although undirected acyclic graphs are limited to trees, DAGs can be considerably more complicated. They just have to avoid directed cycles, as shown in Figure

.

A topological sort of a directed acyclic graph is an ordering on the vertices such that all edges go from left to right. Only an acyclic graph can have a topological sort, because a directed cycle must eventually return home to the source of the cycle. However, every DAG has at least one topological sort, and we can use depth-first search to find such an ordering. Topological sorting proves very useful in scheduling jobs in their proper sequence, as discussed in catalog Section

.

Depth-first search can be used to test whether a graph is a DAG, and if so to find a topological sort for it. A directed graph is a DAG if and only if no back edges are encountered during a depth-first search. Labeling each of the vertices in the reverse order that they are marked completely-explored finds a topological sort of a DAG. Why? Consider what happens to each directed edge

as we encounter it

during the exploration of vertex u: ●

If v is currently undiscovered, then we then start a DFS of v before we can continue with u. Thus v is marked completely-explored before u is, and v appears before u in the topological order, as it must.

●

If v is discovered but not completely-explored, then

●

DAG. If v is completely-explored, then it will have been so labeled before u. Therefore, u appears before v in the topological order, as it must.

file:///E|/BOOK/BOOK2/NODE70.HTM (1 of 2) [19/1/2003 1:29:10]

is a back edge, which is forbidden in a

Topological Sorting

Next: Articulation Vertices Up: Applications of Graph Traversal Previous: Two-Coloring Graphs Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE70.HTM (2 of 2) [19/1/2003 1:29:10]

Articulation Vertices

Next: Modeling Graph Problems Up: Applications of Graph Traversal Previous: Topological Sorting

Articulation Vertices

Figure: An articulation vertex is the weakest point in the graph Suppose you are a terrorist seeking to disrupt the telephone network. Which station in Figure should you choose to blow up to cause the maximum amount of damage? An articulation vertex is a vertex of a connected graph whose deletion disconnects the graph. Any graph that contains an articulation vertex is inherently fragile, because deleting that single vertex causes a loss of connectivity. In general, the connectivity of a graph is the smallest number of vertices whose deletion will disconnect the graph. For graphs with an articulation vertex, the connectivity is one. Connectivity is an important measure of robustness in network design, as discussed in catalog Section

.

A simple application of either depth-first or breadth-first search suffices to find all the articulation vertices in a graph in O(n (m+n)). For each vertex v, delete it and then do a BFS traversal of the remaining graph to establish whether it is still connected. In fact, there is a clever O(n+m) algorithm that tests all the vertices using only a single depth-first search. Additional information on edge and vertex connectivity testing appears in Section

.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE71.HTM [19/1/2003 1:29:11]

Modeling Graph Problems

Next: Minimum Spanning Trees Up: Graph Algorithms Previous: Articulation Vertices

Modeling Graph Problems Proper modeling is the key to making effective use of graph algorithms. We have seen a variety of definitions of graph properties, and algorithms for computing them. All told, about two dozen different graph problems are presented in the catalog, mostly in Sections framework for modeling most applications.

and

. These problems provide a

The applications below demonstrate the power of proper modeling. Each of them arose in a real-world application as stated, and each can be modeled as a graph problem. Some of the modelings are quite clever, but they illustrate the versatility of graphs in representing relationships. As you read the problem, try to devise an appropriate graph representation before peeking to see how we did it. ●

``I'm looking for an algorithm to design natural routes for video-game characters to follow through an obstacle-filled room. How should I do it?'' Presumably the route that is wanted is the path that looks most like the one that an intelligent being would choose. Since intelligent beings are either lazy or efficient, this should be modeled as some kind of shortest path problem. But what is the graph? One approach would be to lay a grid of points in the room and have a vertex for each point that is a valid place for the character to stand, i.e. so it does not lie within an obstacle. There will be an edge between any pair of nearby vertices, weighted according to the distance between them. The shortest path between two vertices will be close to the shortest path between the points. Although direct geometric methods are known for shortest paths (see Section

●

), it is easier to model this discretely as a graph. ``In DNA sequencing, we are given experimental data consisting of small fragments. For each fragment f, we have certain other fragments that are forced to lie to the left of f, certain fragments forced to be to the right of f, and the remaining fragments, which are free to go on either side. How can we find a consistent ordering of the fragments from left to right that satisfies all the constraints?'' Create a directed graph, where each fragment is assigned a unique vertex. Insert a directed edge from any fragment l that is forced to be to the left of f, and a directed edge

file:///E|/BOOK/BOOK2/NODE72.HTM (1 of 3) [19/1/2003 1:29:12]

to any

Modeling Graph Problems

●

●

fragment r forced to be to the right of f. We seek an ordering of the vertices such that all the edges go from left to right. This is exactly a topological sort of the resulting directed acyclic graph. The graph must be acyclic for this to work, because cycles make finding a consistent ordering impossible. ``In my graphics work I need to solve the following problem. Given an arbitrary set of rectangles in the plane, how can I distribute them into a minimum number of buckets such that the subset of rectangles in the same bucket do not intersect each other? In other words, there should not be any overlapping area between any two rectangles in the same bucket.'' We can formulate a graph where each vertex is a rectangle, and there is an edge if two rectangles intersect. Each bucket corresponds to an independent set of rectangles, so there is no overlap between any two. A vertex coloring of a graph is a partition of the vertices into independent sets, so minimizing the number of colors is exactly what you want. ``In porting code from UNIX to DOS, I have to shorten the names of several hundred files down to at most 8 characters each. I can't just take the first eight characters from each name, because ``filename1'' and ``filename2'' will get assigned the exact same name. How can I shorten the names while ensuring that they do not collide?'' Construct a graph with vertices corresponding to each original file name for , as well . Add an edge between each as a collection of acceptable shortenings for each name original and shortened name. Given such a formulation, we seek a set of n edges that have no vertices in common, because the file name of each is thus mapped to a distinct acceptable

●

, is exactly this problem of finding an substitute. Bipartite matching, discussed in Section independent set of edges in a graph. ``In organized tax fraud, criminals submit groups of phony tax returns in the hopes of getting undeserved refunds. These phony returns are all similar, but not identical. How can we detect clusters of similar forms so the IRS can nail the cheaters?'' A natural graph model treats each form as a vertex and adds an edge between any two tax forms that are suspiciously similar. A cluster would correspond to a group of forms with many edges between them. In particular, a clique is a set of k vertices with all possible edges between them.

●

Any sufficiently large clique identifies a cluster worth studying. ``In the optical character-recognition system that we are building, we need a way to separate the lines of text. Although there is some white space between the lines, problems like noise and the tilt of the page makes it hard to find. How can we do line segmentation? Consider the following graph formulation. Treat each pixel in the image as a vertex in the graph, with an edge between two neighboring pixels. The weight of this edge should be proportional to how dark the pixels are. A segmentation between two lines is a path in this graph from the left to right side of the page. Of all possible paths, we seek a relatively straight path that avoids as much blackness as possible. This suggests that the shortest path in the pixel graph will likely find a

file:///E|/BOOK/BOOK2/NODE72.HTM (2 of 3) [19/1/2003 1:29:12]

Modeling Graph Problems

good line segmentation.

Next: Minimum Spanning Trees Up: Graph Algorithms Previous: Articulation Vertices Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE72.HTM (3 of 3) [19/1/2003 1:29:12]

Minimum Spanning Trees

Next: Prim's Algorithm Up: Graph Algorithms Previous: Modeling Graph Problems

Minimum Spanning Trees A tree is a connected graph with no cycles. A spanning tree is a subgraph of G that has the same set of vertices of G and is a tree. A minimum spanning tree of a weighted graph G is the spanning tree of G whose edges sum to minimum weight.

Figure: Two spanning trees of point set (a); the minimum spanning tree (b), and the shortest path from center tree (c) Minimum spanning trees are useful in finding the least amount of wire necessary to connect a group of homes or cities, as illustrated in Figure . In such geometric problems, the point set defines a assigned a weight equal to the distance from to . Additional complete graph, with edge applications of minimum spanning trees are discussed in Section

.

A minimum spanning tree minimizes the total length over all possible spanning trees. However, there can be more than one minimum spanning tree in any graph. Consider a graph G with m identically weighted edges. All spanning trees of G are minimum spanning trees, since each contains exactly n-1 equal-weight edges. For general weighted graphs, however, the problem of finding a minimum spanning tree is more difficult. It can, however, be solved optimally using two different greedy algorithms. Both are presented below, to illustrate how we can demonstrate the optimality of certain greedy heuristics.

file:///E|/BOOK/BOOK2/NODE73.HTM (1 of 2) [19/1/2003 1:29:13]

Minimum Spanning Trees

● ●

Prim's Algorithm Kruskal's Algorithm

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE73.HTM (2 of 2) [19/1/2003 1:29:13]

Prim's Algorithm

Next: Kruskal's Algorithm Up: Minimum Spanning Trees Previous: Minimum Spanning Trees

Prim's Algorithm

Every vertex will appear in the minimum spanning tree of any connected graph G. Prim's minimum spanning tree algorithm starts from one vertex and grows the rest of the tree one edge at a time. In greedy algorithms, we make the decision of what to do next by selecting the best local option from all available choices without regard to the global structure. Since we seek the tree of minimum weight, the natural greedy algorithm for minimum spanning tree repeatedly selects the smallest weight edge that will enlarge the tree.

Figure: Where Prim's algorithm goes bad? No, because

Prim-MST(G)

Select an arbitrary vertex

to start the tree from.

While (there are still non-tree vertices)

Select the edge of minimum weight between a tree and non-tree vertex

Add the selected edge and vertex to the tree

.

Prim's algorithm clearly creates a spanning tree, because no cycle can be introduced by adding edges between tree and non-tree vertices. However, why should it be of minimum weight over all spanning trees? We have seen ample evidence of other natural greedy heuristics that do not yield a global optimium. Therefore, we must be particularly careful to demonstrate any such claim. Suppose that there existed a graph G for which Prim's algorithm did not return a minimum spanning tree. Since we are building the tree incrementally, this means that there must have been some particular instant where we went wrong. Before we file:///E|/BOOK/BOOK2/NODE74.HTM (1 of 2) [19/1/2003 1:29:14]

Prim's Algorithm

inserted edge (x,y), consisted of a set of edges that was a subtree of a minimum spanning tree , but choosing edge (x,y) took us away from a minimum spanning tree. But how could it? There must be a path p from x to y in , using an edge , where is in but is not. This edge must have weight at least that of (x,y), or else Prim's algorithm would have selected it instead of (x,y) when it had the chance. Inserting (x,y) and deleting from leaves a spanning tree no larger than before, meaning that Prim's algorithm could not have made a fatal mistake in selecting edge (x,y). Therefore, by contradiction, Prim's algorithm has to construct a minimum spanning tree. Prim's algorithm is correct, but how efficient is it? That depends on which data structures are used to implement it, but it should be clear that O(nm) time suffices. In each of n iterations, we will scan through all the m edges and test whether the current edge joins a tree with a non-tree vertex and whether this is the smallest edge seen thus far. By maintaining a Boolean flag along with each vertex to denote whether it is in the tree or not, this test can be performed in constant time. In fact, better data structures lead to a faster, , implementation by avoiding the need to sweep through more than n edges in any iteration.

Next: Kruskal's Algorithm Up: Minimum Spanning Trees Previous: Minimum Spanning Trees Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE74.HTM (2 of 2) [19/1/2003 1:29:14]

Kruskal's Algorithm

Next: Shortest Paths Up: Minimum Spanning Trees Previous: Prim's Algorithm

Kruskal's Algorithm

Kruskal's algorithm is an alternative approach to finding minimum spanning trees that is more efficient on sparse graphs. Like Prim's, Kruskal's algorithm is greedy; unlike Prim's, it does not start with a particular vertex. Kruskal's algorithm works by building up connected components of the vertices. Initially, each vertex forms its own separate component in the tree-to-be. The algorithm repeatedly considers the lightest remaining edge and tests whether the two endpoints lie within the same connected component. If so, the edge will be discarded, because adding it will create a cycle in the tree-to-be. If the endpoints are in different components, we insert the edge and merge the components. Since each connected component is always a tree, we need never explicitly test for cycles:

Kruskal-MST(G)

Put the edges in a priority queue ordered by weight.

count=0

while (count < n-1) do

get next edge (v,w)

if (component (v)

component(w))

add to

merge component(v) and component(w)

This algorithm adds n-1 edges without creating a cycle, so clearly it creates a spanning tree of any connected graph. But why must this be a minimum spanning tree? Suppose it wasn't. As with the correctness proof of Prim's algorithm, there must be some graph for which it fails, and in particular there must a single edge (x,y) whose insertion first prevented the tree from being a minimum spanning tree . Inserting edge (x,y) in will create a cycle with the path from x to y. Since x would have been and y were in different components at the time of inserting (x,y), at least one edge on this path considered by Kruskal's algorithm after (x,y) was. But this means that , so exchanging the two edges yields

file:///E|/BOOK/BOOK2/NODE75.HTM (1 of 2) [19/1/2003 1:29:15]

Kruskal's Algorithm

a tree of weight at most

. Therefore, we could not have made a mistake in selecting (x,y), and the correctness follows.

What is the time complexity of Kruskal's algorithm? Inserting and retrieving m edges from a priority queue such as a heap time. The while loop makes at most m iterations, each testing the connectivity of two trees plus an edge. In takes the most simple-minded approach, this can be implemented by a breadth-first or depth-first search in a graph with at most n edges and n vertices, thus yielding an O(mn) algorithm. However, a faster implementation would result if we could implement the component test in faster than O(n) time. In fact, the union-find data structure, discussed in Section , can support such queries in time. With this data structure, Kruskal's algorithm runs in time, which is faster than Prim's for sparse graphs. Observe again the impact that the right data structure can have in implementing a straightforward algorithm.

Next: Shortest Paths Up: Minimum Spanning Trees Previous: Prim's Algorithm Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE75.HTM (2 of 2) [19/1/2003 1:29:15]

Shortest Paths

Next: Dijkstra's Algorithm Up: Graph Algorithms Previous: Kruskal's Algorithm

Shortest Paths

Figure: The shortest path from to t can pass through many intermediate vertices. The shortest path between two vertices and t in an unweighted graph can be constructed using a breadthfirst search from . When we first encounter t in the search, we will have reached it from using the minimum number of possible edges. This minimum-link path is recorded in the breadth-first search tree, and it provides the shortest path when all edges have equal weight. However, in an arbitrary weighted graph, the weight of a path between two vertices is the sum of the weights of the edges along the path. The shortest path might use a large number of edges, just as the shortest route (timewise) from home to office may involve shortcuts using backroads and many turns, as shown in Figure Shortest paths have a surprising variety of applications. See catalog Section Section

● ●

for further examples.

Dijkstra's Algorithm All-Pairs Shortest Path

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE76.HTM [19/1/2003 1:29:16]

.

and the war story of

Dijkstra's Algorithm

Next: All-Pairs Shortest Path Up: Shortest Paths Previous: Shortest Paths

Dijkstra's Algorithm

We can use Dijkstra's algorithm to find the shortest path between any two vertices (,t) in a weighted graph, where each edge has non-negative edge weight. Although most applications of shortest path involve graphs with positive edge weights, such a condition is not needed for either Prim's or Kruskal's algorithm to work correctly. The problems that negative edges cause Dijkstra's algorithm will become apparent once you understand the algorithm. The principle behind Dijkstra's algorithm is that given the shortest path between and each of a given set of vertices , there must exist some other vertex x such that the shortest path from to x must go from to to x, for some . Specifically, it is the vertex x that minimizes over all , where w(i,j) is the length of the edge from i to j and dist(i,j) is the length of the shortest path between them. This suggests a dynamic programming-like strategy. The shortest path from to itself is trivial unless there are negative weight edges, so dist(,)=0. Armed with the shortest path to , if (,y) is the lightest edge incident to , then d(,y) = w(,y). As soon as we decide that we have determined the shortest path to a node x, we search through all the outgoing edges of x to see whether there is a better path from to some unknown vertex through x:

ShortestPath-Dijkstra(G,s,t)

for i=1 to n,

for each edge (, v), dist[v]=w(, v)

last=

while (

)

select

, the unknown vertex minimizing dist[v]

for each edge

file:///E|/BOOK/BOOK2/NODE77.HTM (1 of 2) [19/1/2003 1:29:17]

,

Dijkstra's Algorithm

To be certain of finding the shortest path between and t, we might have to first find the shortest path between and all other vertices. This defines a shortest path spanning tree rooted in . For undirected graphs, this will be the breadth-first search tree, but in general it provides the shortest path between and all other vertices. What is the running time of this algorithm? When implemented using adjacency lists and a Boolean array to mark what is known about each vertex, the complexity is . This is the same running time as a proper version of Prim's algorithm; indeed, except for the extension condition, it is the same algorithm as Prim's.

Next: All-Pairs Shortest Path Up: Shortest Paths Previous: Shortest Paths Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE77.HTM (2 of 2) [19/1/2003 1:29:17]

All-Pairs Shortest Path

Next: War Story: Nothing but Up: Shortest Paths Previous: Dijkstra's Algorithm

All-Pairs Shortest Path

If we want to find the length of the shortest path between all

pairs of vertices, we could run Dijkstra's algorithm n

times, once from each possible starting vertex. This yields a cubic time algorithm for all-pairs shortest path, since . Can we do better? Significantly improving the complexity is still an open question, but there is a superslick dynamic programming algorithm that also runs in . There are several ways to characterize the shortest path between two nodes in a graph. The Floyd-Warshall algorithm starts by numbering the vertices of the graph from 1 to n. We use these numbers here not to label the vertices, but to order them. Define to be the length of the shortest path from i to j using only vertices numbered from 1, 2,..., k as possible intermediate vertices. What does this mean? When k = 0, we are allowed no intermediate vertices, so that every path consists of at most one edge. Thus . In general, adding a new vertex k+1 as a possible intermediary helps only if there is a short path that goes through it, so

This recurrence performs only a constant amount of work per cell. The following dynamic programming algorithm implements the recurrence:

Floyd(G)

Let

, the weight matrix of G

for k=1 to n

for i=1 to n

for j=1 to n

file:///E|/BOOK/BOOK2/NODE78.HTM (1 of 2) [19/1/2003 1:29:18]

All-Pairs Shortest Path

The Floyd-Warshall all-pairs shortest path runs in time, which asymptotically is no better than n calls to Dijkstra's algorithm. However, the loops are so tight and the program so short that it runs better in practice. It is also notable as one of the rare algorithms that work better on adjacency matrices than adjacency lists.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE78.HTM (2 of 2) [19/1/2003 1:29:18]

War Story: Nothing but Nets

Next: War Story: Dialing for Up: Graph Algorithms Previous: All-Pairs Shortest Path

War Story: Nothing but Nets I'd been tipped off about a small printed-circuit board testing company nearby that was in need of some algorithmic consulting. And so I found myself inside a typically non-descript building in a typically nondescript industrial park, talking with the president of Integri-Test, along with one of his lead technical people. ``We're the leaders in robotic printed-circuit board testing devices. Our customers have very high reliability requirements for their PC-boards. They must check that each and every board has no wire breaks before filling it with components. This means testing that each and every pair of points on the board that are supposed to be connected are connected.'' ``How do you do the testing?'' I asked. ``We have a robot with two arms, each with electric probes. To test whether two points are properly connected, the arms simultaneously contact both of the points. If the two points are propertly connected, then the probes will complete a circuit. For each net, we hold one arm fixed at one point and move the other to cover the rest of the points.'' ``Wait!'' I cried. ``What is a net?''

Figure: An sample net showing (a) the metal connection layer, (b) the contact points, (c) their minimum spanning tree, and (d) the points partitioned into clusters

file:///E|/BOOK/BOOK2/NODE79.HTM (1 of 4) [19/1/2003 1:29:19]

War Story: Nothing but Nets

``On a circuit board there are certain sets of points that are all connected together with a metal layer. This is what we mean by a net. Sometimes a net consists of two points, i.e. is an isolated wire. Sometimes a net can have 100 to 200 points, like all the connections to power or ground.'' ``I see. So you have a list of all the connections between pairs of points on the circuit board, and you want to trace out these wires.'' He shook his head. ``Not quite. The input for our testing program consists only of the net contact points, as shown in Figure (b). We don't know where the actual wires are, but we don't have to. All we have to do is verify that all the points in a net are connected together. We do this by putting the left robot arm on the leftmost point in the net, then having the right arm move around to all the other points in the net to test if they are connected to the left point. If they are all connected to the left point, it means that they must all be connected to each other.'' I thought for a moment about what this meant. ``OK. So your right arm has to visit all the other points in the net. How do you choose the order to visit them?'' The technical guy spoke up. ``Well, we sort the points from left to right and then go in that order. Is that a good thing to do?'' ``Have you ever heard of the traveling salesman problem?'' I asked. He was an electrical engineer, not a computer scientist. ``No, what's that?'' he asked. ``Traveling salesman is the name of the exact problem that you are trying to solve. Given a set of points you have to visit, how do you order them so as to minimize the travel time. Algorithms for the traveling salesman problem have been extensively studied. For small nets, by doing an exhaustive search you will be able to find the optimal tour. For big nets, there are heuristics that will get you very close to the optimal tour.'' I would have pointed them to Section

if I had had this book handy.

The president scribbled down some notes and then frowned. ``Fine. Maybe you can order the points in a net better for us. But that is not our real problem. When you watch our robot in action, the right arm sometimes has to run all the way to the right side of the board on a given net, while the left arm just sits there. It seems we would benefit by breaking a net into smaller pieces to balance things out.'' I sat down and thought. The left and right arms were each going to have interlocking TSP problems to solve. The left arm would move between the leftmost points of each net, while the right arm was going to visit all the other points in each net as ordered by the left TSP tour. By breaking each net into smaller nets, so that each net occupies a small chunk of real estate, we would avoid making the right arm cross all the way across the board. Further, a lot of little nets meant there would be more points in the left TSP, so each left-arm movement was likely to be short, too. file:///E|/BOOK/BOOK2/NODE79.HTM (2 of 4) [19/1/2003 1:29:19]

War Story: Nothing but Nets

``You are right. We should win if we can break big nets into small nets. We want the nets to be small, both in the number of points and in the area of the net. But we must be sure that if we validate the connectivity of each small net, we will have confirmed that the big net is connected. Whenever there is one point in common between two little nets, that is enough to show that the bigger net formed by the two little nets is connected, since current can flow between any pair of points.'' Now we had to break each net into overlapping pieces, where each piece was small. This is a clustering . In fact, that problem. Minimum spanning trees are often used for clustering, as discussed in Section was the answer! We could find the minimum spanning tree of the net points and break it into little (d), each cluster would share clusters whenever a spanning tree edge got too long. As shown in Figure exactly one point in common with another cluster, with connectivity ensured because we are covering the edges of a spanning tree. The shape of the clusters would reflect the points in the net, exactly as we would want. If the points lay along a line across the board, the minimum spanning tree would be a path, and the clusters would be pairs of points. If the points all fell in a tight region, there would be one nice fat cluster that the right arm would just scoot around. So I explained the idea of constructing the minimum spanning tree of a graph. The boss listened, scribbled more notes, and frowned again. ``I like your clustering idea. But these minimum spanning trees you talk about are defined on graphs. All you got are points. Where do the weights of the edges come from?'' ``Oh, we can think of it as a complete graph, where every pair of points are connected. The weight of the edge defined by the two points is simply the distance. Or is it...?'' I went back to thinking. The edge cost between two points should reflect the travel time between them. While the distance was related to the travel time, it wasn't necessarily exactly the same thing. ``Hey. I have a question about your robot. Does it take the same amount of time to move the arm leftright as it does up-down?'' They thought a minute. ``Yeah, it does. We use the same type of motors to control horizontal and vertical movements. Further, since the two motors for each arm are independent, we can simultaneously move each arm both horizontally and vertically.'' ``That so? The time to move both one foot left and one foot up is exactly the same as just moving one foot left? This means that the weight cost for each edge in the graph should not be the Euclidean distance between the two points, but the biggest difference between either the x- or y-coordinate. This is something we call the metric, but we can capture it by changing the edge weights in the graph. Anything else funny about your robots?'' I asked.

file:///E|/BOOK/BOOK2/NODE79.HTM (3 of 4) [19/1/2003 1:29:19]

War Story: Nothing but Nets

``Well, it takes some time for the robot to come up to speed. We should probably also factor in acceleration and deceleration of the arms.'' ``Darn right. The more accurately you can model the time your arm takes to move between two points, the better our solution will be. But now we have a very clean formulation. Let's code it up and let's see how well it works!'' They were somewhat skeptical whether this approach would do any good, but they agreed to think about it. A few weeks later they called me back and reported that the new algorithm reduced testing time by about 30% over their previous approach, at a cost of a little more computational preprocessing. However, since their testing machine costs $200,000 a pop and a PC costs $2,000, this is an excellent tradeoff. It is particularly advantageous since the preprocessing need only be done once when testing multiple instances of the same board. The key idea leading to the successful solution was knowing how to model the job in terms of classical algorithmic graph problems. I smelled TSP the instant they started talking about minimizing robot motion. Once I realized that they were implicitly forming a star-shaped spanning tree to ensure connectivity, it was natural to ask whether a minimum spanning tree would perform any better. This idea led to a natural way to think about clustering, and thus partitioning each net into smaller nets. Finally, by carefully constructing our distance metric to accurately model the costs of the robot itself, we get to incorporate quite complicated properties (such as acceleration and differences between horizontal and vertical speeds) without changing our fundamental graph model or algorithm design.

Next: War Story: Dialing for Up: Graph Algorithms Previous: All-Pairs Shortest Path Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE79.HTM (4 of 4) [19/1/2003 1:29:19]

War Story: Dialing for Documents

Next: Exercises Up: Graph Algorithms Previous: War Story: Nothing but

War Story: Dialing for Documents I was part of a group visiting Periphonics, an industry leader in building telephone voice-response systems. These are more advanced versions of the Press 1 for more options, Press 2 if you didn't press 1 telephone systems that have come to blight everyone's lives in recent years. We were being given the standard tour when someone from our group asked, ``Why don't you guys use voice recognition for data entry. It would be a lot less annoying than typing things out on the keypad.'' The tour guide reacted smoothly. ``Our customers have the option of incorporating speech recognition into our products, but very few of them do. User-independent, connected-speech recognition is not accurate for most applications. Our customers prefer building systems around typing text on the telephone keyboards.'' ``Prefer typing, my pupik!'', came a voice from the rear of our group. ``I hate typing on a telephone. Whenever I call my brokerage house to get stock quotes, I end up talking to some machine, which asks me to type in the three letter code. To make it worse, I have to hit two buttons to type in one letter, in order to distinguish between the three letters printed on each key of the telephone. I hit the 2 key and it says Press 1 for `A', Press 2 for `B', Press 3 for `C'. Pain the neck if you ask me.'' ``Maybe you don't really have to hit two keys for each letter?'' I chimed in. ``Maybe the system could figure out the correct letter from context?'' ``There isn't a whole lot of context when you type in three letters of stock market code.'' ``Sure, but there would be plenty of context if we were typing in English sentences. I'll bet that we could reconstruct English sentences correctly if they were typed in a telephone at one keystroke per letter.'' The guy from Periphonics gave me a disinterested look, then continued the tour. But when I got back to the office, I decided to give it a try. It was clear that not all letters were equally likely to be typed on a telephone. In fact, not all letters can be typed, since `Q' and `Z' are not labeled on the standard American telephone. Therefore, we adopted the convention that `Q', `Z', and space were all on the * key. We could take advantage of the uneven file:///E|/BOOK/BOOK2/NODE80.HTM (1 of 6) [19/1/2003 1:29:21]

War Story: Dialing for Documents

distribution of letter frequencies to help us decode the text. For example, if you hit the 3 key while typing English, you were more likely to have meant to type an `E' than either a `D' or `F'. By taking into account the frequencies of a window of three characters, we could predict the typed text. Indeed, this is what happened when I tried it on the Gettysburg Address: enurraore ane reten yeasr ain our ectherr arotght eosti on ugis aootinent a oey oation aoncdivee in licesty ane eedicatee un uhe rrorosition uiat all oen are arectee e ual ony ye are enichde in a irect aitil yar uestini yhethes uiat oatioo or aoy oation ro aoncdivee ane ro eedicatee aan loni eneure ye are oet on a irect aattlediele oe uiat yar ye iate aone un eedicate a rostion oe uiat eiele ar a einal restini rlace eor uiore yin iere iate uhdis lives uiat uhe oation ogght live it is aluniethes eittini ane rrores uiat ye rioule en ugir att in a laries reore ye aan oou eedicate ye aan oou aoorearate ye aan oou ialloy ugis iroune the arate oen litini ane eeae yin rustgilee iere iate aoorearatee it ear aante our roor rowes un ade or eeuraat the yople yill little oote oor loni renences yiat ye ray iere att it aan oetes eosiet yiat uhfy eie iere it is eor ur uhe litini rathes un ae eedicatee iere un uhe undiniside yopl yhici uhfy yin entght iere iate uiur ear ro onaky aetancde it is rathes eor ur un ae iere eedicatee un uhe irect uarl rencinini adeore ur uiat eron uhere ioooree eeae ye uale inarearee eeuotion uo tiat aaure eor yhici uhfy iere iate uhe lart eull oearure oe eeuotioo tiat ye iere iggily rerolue uiat uhere eeae riall oou iate eide io

The trigram statistics did a decent job of translating it into Greek, but a terrible job of transcribing English. One reason was clear. This algorithm knew nothing about English words. If we coupled it with a dictionary, we might be on to something. The difficulty was that often two words in the dictionary would be represented by the exact same string of phone codes. For an extreme example, the code string ``22737'' collides with eleven distinct English words, including cases, cares, cards, capes, caper, and bases. As a first attempt, we reported the unambiguous characters of any words that collided in the dictionary, and used trigrams to fill in the rest of the characters. We were rewarded with: eourscore and seven yearr ain our eatherr brought forth on this continent azoey nation conceivee in liberty and dedicatee uo uhe proposition that all men are createe equal ony ye are engagee in azipeat civil yar uestioi whether that nation or aoy nation ro conceivee and ro dedicatee aan long endure ye are oet on azipeat battlefield oe that yar ye iate aone uo dedicate a rostion oe that field ar a final perthni place for those yin here iate their lives that uhe nation oight live it is altogether fittinizane proper that ye should en this aut in a larges sense ye aan oou dedicate ye aan oou consecrate ye aan oou hallow this ground the arate men litioi and deae yin strugglee here iate consecratee it ear above our roor power uo ade or detract the world will little oote oor long remember what ye ray here aut it aan meter forget what uhfy die here it is for ur uhe litioi rather uo ae dedicatee here uo uhe toeioisgee york which uhfy yin fought here iate thus ear ro mocky advancee it is rather for ur uo ae here dedicatee uo uhe great task renagogoi adfore ur that from there honoree deae ye uale increasee devotion uo that aause for which uhfy here iate uhe last eull measure oe devotion that ye here highky resolve that there deae shall oou iate fide io vain that this nation under ioe shall iate azoey birth oe freedom and that ioternmenu oe uhe people ay uhe people for uhe people shall oou perish from uhe earth

file:///E|/BOOK/BOOK2/NODE80.HTM (2 of 6) [19/1/2003 1:29:21]

War Story: Dialing for Documents

If you were a student of American history, maybe you could recognize it, but you certainly couldn't read it. Somehow we had to distinguish between the different dictionary words that got hashed to the same code. We could factor in the relative popularity of each word, which would help, but this would still make too many mistakes. At this point I started working with Harald Rau on the project, who proved a great collaborator for two reasons. First, he was a bright and peristent graduate student. Second, as a native German speaker he would believe every lie I told him about English grammar.

Figure: The phases of the telephone code reconstruction process Harald built up a phone code reconstruction program on the lines of Figure . It worked on the input one sentence at a time, identifying dictionary words that matched each code string. The key problem was how to incorporate the grammatical constraints. ``We can get good word-use frequencies and grammatical information using this big text database called the Brown Corpus. It contains thousands of typical English sentences, each of which is parsed according to parts of speech. But how do we factor it all in?'' Harald asked. ``Let's try to think about it as a graph problem,'' I suggested. file:///E|/BOOK/BOOK2/NODE80.HTM (3 of 6) [19/1/2003 1:29:21]

War Story: Dialing for Documents

``Graph problem? What graph problem? Where is there even a graph?'' ``Think of a sentence as a list of phone tokens, each representing a word in the sentence. For each phone token, we have a list of words from the dictionary that match it. How can we choose which one is right? Each possible sentence interpretation can be thought of as a path in a graph. The vertices of this graph will be the complete set of possible word choices. There will be an edge from a possible choice for the ith word to each possible choice for the (i+1)st word. The cheapest path across this graph is the right interpretation of the sentence.''

Figure: The minimum-cost path through the graph defines the best interpretation for a sentence ``But all the paths look the same. They have the same number of edges. Wait. Now I see! To make the paths different, we have to weight the edges.'' ``Exactly! The cost of an edge will reflect how likely it is that we will want to travel through the given pair of words. Maybe we can count how often that pair of words occurred together in previous texts. Or we can weight by what part of speech each word is. Maybe nouns don't like to be next to nouns as much as they like being next to verbs.'' ``It will be hard to keep track of word-pair statistics, since there are so many of them. But we certainly know the frequency of each word. How can we factor that into things?'' ``We can pay a cost for walking through a particular vertex that depends upon the frequency of the word. Our best sentence will be given by the shortest path across the graph.'' ``But how do we figure out the relative weights of these factors?'' ``Well, try what seems natural to you and then we can experiment with it.'' Harald incorported this shortest-path algorithm. With proper grammatical and statistical constraints, the system performed great, as reported in [RS96]. Look at the Gettysburg Address now, with all the reconstruction errors highlighted: FOURSCORE AND SEVEN YEARS AGO OUR FATHERS BROUGHT FORTH ON THIS

file:///E|/BOOK/BOOK2/NODE80.HTM (4 of 6) [19/1/2003 1:29:21]

War Story: Dialing for Documents

CONTINENT A NEW NATION CONCEIVED IN LIBERTY AND DEDICATED TO THE PROPOSITION THAT ALL MEN ARE CREATED EQUAL. NOW WE ARE ENGAGED IN A GREAT CIVIL WAR TESTING WHETHER THAT NATION OR ANY NATION SO CONCEIVED AND SO DEDICATED CAN LONG ENDURE. WE ARE MET ON A GREAT BATTLEFIELD OF THAT WAS. WE HAVE COME TO DEDICATE A PORTION OF THAT FIELD AS A FINAL SERVING PLACE FOR THOSE WHO HERE HAVE THEIR LIVES THAT THE NATION MIGHT LIVE. IT IS ALTOGETHER FITTING AND PROPER THAT WE SHOULD DO THIS. BUT IN A LARGER SENSE WE CAN NOT DEDICATE WE CAN NOT CONSECRATE WE CAN NOT HALLOW THIS GROUND. THE BRAVE MEN LIVING AND DEAD WHO STRUGGLED HERE HAVE CONSECRATED IT FAR ABOVE OUR POOR POWER TO ADD OR DETRACT. THE WORLD WILL LITTLE NOTE NOR LONG REMEMBER WHAT WE SAY HERE BUT IT CAN NEVER FORGET WHAT THEY DID HERE. IT IS FOR US THE LIVING RATHER TO BE DEDICATED HERE TO THE UNFINISHED WORK WHICH THEY WHO FOUGHT HERE HAVE THUS FAR SO NOBLY ADVANCED. IT IS RATHER FOR US TO BE HERE DEDICATED TO THE GREAT TASK REMAINING BEFORE US THAT FROM THESE HONORED DEAD WE TAKE INCREASED DEVOTION TO THAT CAUSE FOR WHICH THEY HERE HAVE THE LAST FULL MEASURE OF DEVOTION THAT WE HERE HIGHLY RESOLVE THAT THESE DEAD SHALL NOT HAVE DIED IN VAIN THAT THIS NATION UNDER GOD SHALL HAVE A NEW BIRTH OF FREEDOM AND THAT GOVERNMENT OF THE PEOPLE BY THE PEOPLE FOR THE PEOPLE SHALL NOT PERISH FROM THE EARTH.

Figure: Our telephone-code reconstruction system applied to various text samples While we still made a few mistakes, the results are clearly good enough for a variety of applications. Periphonics certainly thought so, for they later licensed our program to incorporate into their products. Figure shows that we were able to reconstruct over 99% of the characters correctly on a megabyte of President Clinton's speeches, so if Bill had phoned them in, we would certainly still be able to understand it. The reconstruction time is fast enough, indeed faster than you can type it in on the phone keypad. The constraints associated with many different pattern recognition problems can be formulated as shortest path problems in graphs. In fact, there is a particularly convenient dynamic programming solution for these problems known as the Viterbi algorithm, which is used in speech and handwriting recognition systems. Despite the fancy name, all the Viterbi algorithm is doing is solving a shortest path problem. Hunting for a graph formulation for any given problem is always a good way to proceed.

file:///E|/BOOK/BOOK2/NODE80.HTM (5 of 6) [19/1/2003 1:29:21]

War Story: Dialing for Documents

Next: Exercises Up: Graph Algorithms Previous: War Story: Nothing but Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE80.HTM (6 of 6) [19/1/2003 1:29:21]

Exercises

Next: Implementation Challenges Up: Graph Algorithms Previous: War Story: Dialing for

Exercises 1. Present correct and efficient algorithms to convert between the following graph data structures, for an undirected graph G with n vertices and m edges. You must give the time complexity of each algorithm.

2. 3.

4.

5.

1. Convert from an adjacency matrix to adjacency lists. 2. Convert from an adjacency list to an incidence matrix. An incidence matrix M has a row for each vertex and a column for each edge, such that M[i,j]=1 if vertex i is part of edge j, otherwise M[i,j] = 0. 3. Convert from an incidence matrix to adjacency lists. Is the path between a pair of vertices in a minimum spanning tree necessarily a shortest path between the two vertices in the full graph? Give a proof or a counterexample. Assume that all edges in the graph have distinct edge weights (i.e. no pair of edges have the same weight). Is the path between a pair of vertices in a minimum spanning tree necessarily a shortest path between the two vertices in the full graph? Give a proof or a counterexample. Suppose G is a connected undirected graph. An edge e whose removal disconnects the graph is called a bridge. Must every bridge e be an edge in a depth-first search tree of G, or can e be a back edge? Give a proof or a counterexample. (*) In breadth-first and depth-first search, an undiscovered node is marked discovered when it is first encountered, and marked completely-explored when it has been completely searched. At any given moment, several nodes might be simultaneously in the discovered state. (a) Describe a graph on n vertices and a particular starting vertex v such that during a breadth-first search starting from v, nodes are simultaneously in the discovered state. (b) Describe a graph on n vertices and a particular starting vertex v such that during a depth-first nodes are simultaneously in the discovered state. search starting from v,

(c) Describe a graph on n vertices and a particular starting vertex v such that at some point during nodes remain undiscovered, while nodes have been a depth-first search starting from v, completely-explored. (Note, there may also be discovered nodes.) 6. Given the pre-order and in-order traversals of a binary tree, is it possible to reconstruct the tree? If file:///E|/BOOK/BOOK2/NODE81.HTM (1 of 3) [19/1/2003 1:29:23]

Exercises

so, sketch an algorithm to do it. If not, give a counterexample. Repeat the problem if you are given the pre-order and post-order traversals. 7. Suppose an arithmetic expression is given as a tree. Each leaf is an integer and each internal node is one of the standard arithmetical operations (+,-,*,/). For example, the expression 2+3*4+(3*4)/5 could be represented by the tree in Figure

(a).

Figure: Expression 2+3*4+(3*4)/5 as a tree and a DAG. Give an O(n) algorithm for evaluating such an expression, where there are n nodes in the tree. 8. (*) Suppose an arithmetic expression is given as a DAG (directed acyclic graph) with common subexpressions removed. Each leaf is an integer and each internal node is one of the standard arithmetical operations (+,-,*,/). For example, the expression 2+3*4+(3*4)/5 could be represented

9.

10.

11. 12.

13.

14.

(b). Give an O(n+m) algorithm for evaluating such a DAG, where there by the DAG in Figure are n nodes and m edges in the DAG. Hint: modify an algorithm for the tree case to achieve the desired efficiency. (*) Given an undirected graph G with n vertices and m edges, and an integer k, give an O(m+n) algorithm that finds the maximum induced subgraph H of G such that each vertex in H has degree , or prove that no such graph exists. An induced subgraph F=(U,R) of a graph G=(V,E) is a subset of U of the vertices V of G, and all edges R of G such that both vertices of each edge are in U. (*) An articulation vertex of a graph G is a vertex whose deletion disconnects G. Let G be a graph with n vertices and m edges. Give a simple O(n+m) algorithm for finding a vertex of G that is not an articulation vertex, i.e. whose deletion does not disconnect G. (*) Following up on the previous problem, give an O(n+m) algorithm that finds a deletion order for the n vertices such that no deletion disconnects the graph. (Hint: think DFS/BFS.) (*) Let G be a weighted directed graph with n vertices and m edges, where all edges have positive weight. A directed cycle is a directed path that starts and ends at the same vertex and contains at algorithm to find a directed cycle in G of minimum total weight. least one edge. Give an algorithm. Partial credit will be given for an (*) Suppose we are given the minimum spanning tree T of a given graph G (with n vertices and m edges) and a new edge e=(u,v) of weight w that we will add to G. Give an efficient algorithm to find the minimum spanning tree of the graph G + e. Your algorithm should run in O(n) time to receive full credit, although slower but correct algorithms will receive partial credit. (*) (a) Let T be a minimum spanning tree of a weighted graph G. Construct a new graph G' by adding a weight of k to every edge of G. Do the edges of T form a minimum spanning tree of G'?

file:///E|/BOOK/BOOK2/NODE81.HTM (2 of 3) [19/1/2003 1:29:23]

Exercises

Prove the statement or give a counterexample. (b) Let describe a shortest weighted path between vertices and t of a weighted graph G. Construct a new graph G' by adding a weight of k to every edge of G. Does P describe a shortest path from to t in G'? Prove the statement or give a counterexample. 15. (*) In certain graph problems, vertices have can have weights instead of or in addition to the the cost of the edge (x,y). This problem weights of edges. Let be the cost of vertex v, and is concerned with finding the cheapest path between vertices a and b in a graph G. The cost of a path is the sum of the costs of the edges and vertices encountered on the path. ❍

❍

❍

Suppose that each edge in the graph has a weight of zero (while non-edges have a cost of ). Assume that for all vertices (i.e. all vertices have the same cost). Give an efficient algorithm to find the cheapest path from a to b and its time complexity. For partial credit, give a less efficient but correct algorithm. Now suppose that the vertex costs are not constant (but are all positive) and the edge costs remain as above. Give an efficient algorithm to find the cheapest path from a to b and its time complexity. For partial credit, give a less efficient but correct algorithm. Now suppose that both the edge and vertex costs are not constant (but are all positive). Give an efficient algorithm to find the cheapest path from a to b and its time complexity. For partial credit, give a less efficient but correct algorithm.

16. (*) Devise and analyze an algorithm that takes a weighted graph G and finds the smallest change in the cost of a non-MST edge that causes a change in the minimum spanning tree of G. Your algorithm must be correct and run in polynomial time. 17. An arborescence of a directed graph G is a rooted tree such that there is a directed path from the root to every other vertex in the graph. Give an efficient and correct algorithm to test whether G contains an arborescence, and its time complexity. describes an algorithm for constructing the dual graph of the 18. (**) The war story of Section triangulation efficiently, although it does not guarantee linear time. Give a worst-case linear algorithm for the problem.

Next: Implementation Challenges Up: Graph Algorithms Previous: War Story: Dialing for Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE81.HTM (3 of 3) [19/1/2003 1:29:23]

Backtracking

Next: Constructing All Subsets Up: Combinatorial Search and Heuristic Previous: Combinatorial Search and Heuristic

Backtracking Backtracking is a systematic way to go through all the possible configurations of a space. These configurations may be all possible arrangements of objects (permutations) or all possible ways of building a collection of them (subsets). Other applications may demand enumerating all spanning trees of a graph, all paths between two vertices, or all possible ways to partition the vertices into color classes. What these problems have in common is that we must generate each one of the possible configurations exactly once. Avoiding both repetitions and missing configurations means that we must define a systematic generation order among the possible configurations. In combinatorial search, we represent our configurations by a vector , where each element is selected from an ordered set of possible candidates for position i. As shown below, this representation is general enough to encode most any type of combinatorial object naturally. The search procedure works by growing solutions one element at a time. At each step in the search, we will have constructed a partial solution with elements fixed for the first k elements of the vector, where . From this partial , we will construct the set of possible candidates for the (k+1)st position. We will then try to solution extend the partial solution by adding the next element from . So long as the extension yields a longer partial solution, we continue to try to extend it. However, at some point, might be empty, meaning that there is no legal way to extend the current partial solution. If so, we must backtrack, and replace , the last item in the solution value, with the next candidate in . It is this backtracking step that gives the procedure its name:

Backtrack(A)

Compute

, the set of candidate first elements of solution A.

k = 1

while k > 0 do

while

do

(*advance*)

= the next element from

file:///E|/BOOK/BOOK2/NODE84.HTM (1 of 3) [19/1/2003 1:29:24]

Backtracking

if

is a solution,

report it.

k = k + 1

compute

, the set of candidate

kth elements of solution A.

k = k - 1

(*backtrack*)

Backtracking constructs a tree of partial solutions, where each vertex is a partial solution. There is an edge from x to y if node y was created by advancing from x. This tree of partial solutions provides an alternative way to think about backtracking, for the process of constructing the solutions corresponds exactly to doing a depth-first traversal of the backtrack tree. Viewing backtracking as depth-first search yields a natural recursive implementation of the basic algorithm:

Backtrack-DFS(A,k)

if

is a solution, report it.

else

k = k +1

compute

while

do

= an element in

=

file:///E|/BOOK/BOOK2/NODE84.HTM (2 of 3) [19/1/2003 1:29:24]

Backtracking

Backtrack(a,k)

Although a breadth-first search could also be used to enumerate all solutions, depth-first search is greatly preferred because of the amount of storage required. In depth-first search, the current state of the search is completely represented by the path from the root to the current search node, which requires space proportional to the height of the tree. In breadth-first search, the queue stores all the nodes at the current level, which is proportional to the width of the search tree. For most interesting problems, the width of the tree will grow exponentially in its height. To really understand how backtracking works, you must see how such objects as permutations and subsets can be constructed by defining the right state spaces. Examples of several state spaces are described below.

● ● ●

Constructing All Subsets Constructing All Permutations Constructing All Paths in a Graph

Next: Constructing All Subsets Up: Combinatorial Search and Heuristic Previous: Combinatorial Search and Heuristic Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE84.HTM (3 of 3) [19/1/2003 1:29:24]

Constructing All Subsets

Next: Constructing All Permutations Up: Backtracking Previous: Backtracking

Constructing All Subsets To design a suitable state space for representing a collection of combinatorial objects, it is important to know how many objects you will need to represent. How many subsets are there of an n-element set, say the integers ? There are exactly two subsets for n=1 namely and , four subsets for n=2, and eight subsets for n=3. Since each new element doubles the number of possibilities, there are subsets of n elements. Each subset is described by stating which elements are in it. Therefore, to construct all subsets, we can set up an array/vector of n cells, where the value of is either true or false, signifying whether the ith item is or is not in the given subset. To use the notation of the general backtrack algorithm, , and A is a solution whenever . Using this state space representation, the backtracking algorithm constructs the following sequence of partial solutions in finding the subsets of . Final solutions, i.e. complete subsets, are marked with a *. False choices correspond to dashes in the partial solution, while true in position i is denoted by i itself:

Trace through this example carefully to make sure you understand the backtracking procedure. The problem of generating subsets is more thoroughly discussed in Section

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE85.HTM [19/1/2003 1:29:25]

.

Constructing All Permutations

Next: Constructing All Paths in Up: Backtracking Previous: Constructing All Subsets

Constructing All Permutations To design a suitable state space for representing permutations, we start by counting them. There are n distinct choices for the value of the first element of a permutation of . Once we have fixed this value of , there are n-1 candidates remaining for the second position, since we can have any value distinct except (repetitions are forbidden). Repeating this argument yields a total of permutations. This counting argument suggests a suitable representation. To construct all n! permutations, set up an array/vector A of n cells. The set of candidates for the ith position will be the set of elements that have not appeared in the (i-1) elements of the partial solution, corresponding to the first i-1 elements of the permutation. To use the notation of the general backtrack algorithm, . The vector A in contains a full solution whenever k = n+1. This representation generates the permutations of the following order:

The problem of generating permutations is more thoroughly discussed in Section

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE86.HTM [19/1/2003 1:29:26]

.

Constructing All Paths in a Graph

Next: Search Pruning Up: Backtracking Previous: Constructing All Permutations

Constructing All Paths in a Graph Enumerating all the simple paths from to t through a given graph is a somewhat more complicated problem than listing permutations or subsets. Unlike the earlier problems, there is no explicit formula that counts the number of solutions as a function of the number of edges or vertices, because the number of paths depends upon the structure of the graph.

Figure: The search tree enumerating all simple paths from vertex 1 in the graph Since the starting point of any path from to t is always , must be . The set of possible candidates for the second position are the vertices v such that (,v) is an edge of the graph, for the path wanders from consists of the set of vertices vertex to vertex using edges to define the legal steps. In general, adjacent to that have not been used in the partial solution A. We can report a successful path whenever . The solution vector A must have room for all n vertices, although most paths are likely to be shorter than this. Figure graph.

shows the search tree giving all paths from a particular vertex in an example

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE87.HTM [19/1/2003 1:29:26]

Search Pruning

Next: Bandwidth Minimization Up: Combinatorial Search and Heuristic Previous: Constructing All Paths in

Search Pruning Backtracking ensures correctness by enumerating all possibilities. For example, a correct algorithm to find the optimal traveling salesman tour could enumerate all n! permutations of n vertices of the graph and selecting the best one. For each permutation, we could check whether each of the n edges implied in the tour really exists in the graph G, and if so, sum the weights of these edges together. For most graphs, however, it would be pointless to construct all the permutations first and then analyze them later. Suppose we started our search from vertex , and it happened that edge was not in G. Enumerating all the (n-2)! permutations beginning with would be a complete waste of effort. Much and continue next with . By carefully restricting the better would be to prune the search after set of next elements to reflect only the moves that are legal from the current partial configuration, we reduce the search complexity significantly. Pruning is the technique of cutting off search the instant we have established that this partial solution cannot be extended into the solution that we want. For example, in our traveling salesman search program, we seek the cheapest tour that visits all vertices before returning to its starting position. Suppose that in the course of our search we find a tour t whose cost is . As the search continues, , where k < n and the sum of the edges on this partial tour perhaps we will find a partial solution . Can there be any reason to continue exploring this node any further? No, assuming all edges is will have cost greater than tour t, and hence have positive cost, because any tour with the prefix is doomed to be non-optimal. Cutting away such failed partial tours as soon as possible can have an enormous impact on running time. Exploiting symmetry is a third avenue for reducing combinatorial search. It is clearly wasteful to evaluate the same candidate solution more than once, because we will get the exact same answer each time we consider it. Pruning away partial solutions identical to those previously considered requires recognizing underlying symmetries in the search space. For example, consider the state of our search for an optimal TSP tour after we have tried all partial positions beginning with . Can it pay to continue the search with partial solutions beginning with ? No. Any tour starting and ending at can be viewed as starting and ending at or any other vertex, for these tours are cycles. There are thus only (n-1)! distinct tours on n vertices, not n!. By restricting the first element of the tour to always be , we save a factor of n in time without missing any interesting solutions. Detecting such symmetries can be subtle, but once identified they can usually be easily exploited by a search program. file:///E|/BOOK/BOOK2/NODE88.HTM (1 of 2) [19/1/2003 1:29:27]

Search Pruning

Next: Bandwidth Minimization Up: Combinatorial Search and Heuristic Previous: Constructing All Paths in Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE88.HTM (2 of 2) [19/1/2003 1:29:27]

Bandwidth Minimization

Next: War Story: Covering Chessboards Up: Combinatorial Search and Heuristic Previous: Search Pruning

Bandwidth Minimization

Figure: A pretty bandwidth-4 layout of a binary tree atop an ugly bandwidth-3 layout To better demonstrate the power of pruning and symmetry detection, let's apply these ideas to producing a search program that solves the bandwidth minimization problem, discussed in detail in catalog Section . I annually run competitions for the fastest bandwidth-minimization program for students in my algorithms courses; the timings below are drawn from these experiences. The bandwidth problem takes as input a graph G, with n vertices and m edges. The goal is to find a gives permutation of the vertices on the line that minimizes the maximum length of any edge. Figure two distinct layouts of a complete binary tree on 15 vertices. The clean, neat layout on the top has a longest edge of length 4, but the seemingly cramped layout on the bottom realizes the optimal bandwidth of 3. The bandwidth problem has a variety of applications, including circuit layout, linear algebra, and optimizing memory usage in hypertext documents. The problem is NP-complete, which implies that no polynomial time worst-case algorithm is known for the problem. It remains NP-complete even for very restricted classes of trees. Since the bandwidth problem seeks a particular permutation, a backtracking program that iterates through all the n! possible permutations and computes the length of the longest edge for each gives a straightforward algorithm. Depending upon how well it is programmed, and how fast a machine it is running on, such an algorithm can expect to solve instances of approximately 8 to 12 vertices within one CPU minute. To speed up this search, we can try to exploit symmetry. For any permutation p, its reverse permutation will realize the exact same bandwidth, since the length of each edge is the same. Hence we can file:///E|/BOOK/BOOK2/NODE89.HTM (1 of 3) [19/1/2003 1:29:29]

Bandwidth Minimization

immediately eliminate half of our search space. The reverse copies are easily removed by placing the leftmost and rightmost elements of the permutation as the first two elements of the vector and then pruning if . Because we are dealing with an exponential search, removing a single factor of two can only be of limited usefulness. Such symmetry elimination might add one to the size of the problem we can do within one CPU minute. For more serious speedups, we need to prune partial solutions. Say we have found a permutation p that yields a longest edge of . By definition, . Suppose that among the elements in a partial layout , where k < n, there is an edge that is at least in length. Can this partial solution expand to provide a better bandwidth solution? Of course not! By pruning the search the instant we have created a long edge, typical instances of 15 to 20 vertices can be solved in one CPU minute, thus providing a substantial improvement. Efforts to further improve the search algorithm must strive for even greater pruning. By using a heuristic method to find the best solution we can before starting to search, we save time by realizing early length cutoffs. In fact, most of the effort in a combinatorial search is typically spent after the optimal solution is found, in the course of proving that no better answer exists. By observing that the optimal bandwidth solution must always be at least half the degree of any vertex (think about the incident edges), we have a lower bound on the size of the optimal solution. We can terminate search soon as we find a solution matching the lower bound. One limitation of this pruning strategy is that only partial solutions of length >b can be pruned, where b is the bandwidth of the best solution to date, since we must place b+1 vertices before we can generate any edges of length at least b. To achieve earlier cutoffs, we can alternately fill in the leftmost and rightmost slots of the configuration, instead of always proceeding from the left. This way, whenever there is an edge between a vertex on the left side and a vertex on the right side, this edge is likely long enough to achieve a cutoff. Pruning can easily occur while positioning the second vertex in the solution vector. Using these enhancements, top-notch programs are capable of solving typical problems on up to 30 vertices consistently within one CPU minute, operating literally millions of times faster than unpruned, untuned efforts. The speed difference between the final and initial versions of the program dwarf the difference between a supercomputer and a microcomputer. Clever search algorithms can easily have a bigger impact on performance than expensive hardware.

Next: War Story: Covering Chessboards Up: Combinatorial Search and Heuristic Previous: Search Pruning

file:///E|/BOOK/BOOK2/NODE89.HTM (2 of 3) [19/1/2003 1:29:29]

Bandwidth Minimization

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE89.HTM (3 of 3) [19/1/2003 1:29:29]

War Story: Covering Chessboards

Next: Heuristic Methods Up: Combinatorial Search and Heuristic Previous: Bandwidth Minimization

War Story: Covering Chessboards Every researcher dreams of solving a classical problem, one that has remained open and unsolved for over a hundred years. There is something romantic about communicating across the generations, being part of the evolution of science, helping to climb another rung up the ladder of human progress. There is also a pleasant sense of smugness that comes from figuring out how to do something that nobody else could do before you. There are several possible reasons why a problem might stay open for such a long period of time. Perhaps the problem is so difficult and profound that it requires a uniquely powerful intellect to solve. A second reason is technological - the ideas or techniques required to solve the problem may not have existed when the problem was first posed. A final possibility is that no one may have cared enough about the problem in the interim to seriously bother with it. Once, I was involved in solving a problem that had been open for over a hundred years. Decide for yourself which reason best explains why. Chess is a game that has fascinated mankind for thousands of years. In addition, it has inspired a number of combinatorial problems of independent interest. The combinatorial explosion was first recognized in the legend that the inventor of chess demanded as payment one grain of rice for the first square of the board, and twice the amount of the ith square for the (i+1)st square, for a total of 36,893,488,147,419,103,231 grains. In beheading him, the wise king first established pruning as a technique for dealing with the combinatorial explosion. In 1849, Kling posed the question of whether all 64 squares on the board can be simultaneously threatened by an arrangement of the eight main pieces on the chess board - the king, queen, two knights, two rooks, and two oppositely colored bishops. Configurations that simultaneously threaten 63 squares have been known for a long time, but whether this was the best possible remained an open problem. This problem seemed ripe for solution by exhaustive combinatorial searching, although whether it was solvable would depend upon the size of the search space. Consider the 8 main pieces in chess (king, queen, two rooks, two bishops, two knights). How many ways can they be positioned on a chessboard? The trivial bound is positions would be unreasonable to search on a modest positions. Anything much larger than about computer in a modest amount of time. Getting the job done would require significant pruning. The first idea is to remove symmetries. file:///E|/BOOK/BOOK2/NODE90.HTM (1 of 3) [19/1/2003 1:29:30]

War Story: Covering Chessboards

Considering the orthogonal and diagonal symmetries, there are only ten distinct positions for the queen. Once the queen is placed, there are 2,080 distinct ways to position a pair of rooks or knights, 64 places to locate the king, and 32 spots for each of the white and black bishops. Thus to perform an exhaustive search, we must test 2,835,349,504,000 distinct positions, still much too large to try. We could use backtracking to construct all of the positions, but we had to find a way to prune the search space significantly if we could hope to finish in our lifetime. Pruning the search meant that we needed a quick way to prove, for a partially filled-in position, that there was no possible way to complete it so as to cover all 64 squares. Suppose we had already placed seven pieces on the board, and together they covered all but 10 squares of the board. Say the remaining piece was the king. Is there any possible position to place the king so that all squares are threatened? The answer must be no, because the king can threaten at most eight squares according to the rules of chess. There can be no reason to bother testing any of the subsequent positions. By pruning these positions, we might win big. Optimizing this pruning strategy required carefully ordering the evaluation of the pieces. Each piece could threaten a certain maximum number of squares: the queen 27, the king 8, the rook 14, and the bishop 13. To maximize the chances of a cutoff, we would want to insert the pieces in decreasing order of mobility. Whenever the number of unthreatened squares exceeds the sum of the maximum coverage of the unplaced pieces, we can prune. This sum is minimized by using the decreasing order of mobility. When we implemented backtrack search with this pruning strategy, we found that it eliminated over of the search space. After optimizing our move generation, our program could search over 1,000 seconds meant 1,000 days! Although positions per second. But this was still too slow, for we might further tweak the program to speed it up by an order of magnitude or so, what we really needed was to find a way to prune more nodes. Effective pruning meant eliminating large numbers of positions at a single stroke. Our previous attempt was too weak. What if instead of placing up to eight pieces on the board simultaneously, we placed more than eight pieces. Obviously, the more pieces we placed simultaneously, the less likely it would be that they didn't threaten all 64 squares. But if they didn't cover, all subsets of eight distinct pieces from the set couldn't possibly threaten all squares. The potential existed to eliminate a vast number of positions by pruning a single node. Thus the nodes of our search tree corresponded to chessboards that could have any number of pieces, and more than one piece on a square. For a given board, we would distinguish two kinds of attack on a square: strong and weak. The notion of strong attack corresponds to the usual notion of attack in chess. A square is weakly attacked if the square is strongly attacked by some subset of the board, that is, weak attack ignores any possible blocking effects of intervening pieces. All 64 squares can be weakly attacked with eight pieces. Our algorithm consists of two passes. The first pass lists all boards such that every square is weakly file:///E|/BOOK/BOOK2/NODE90.HTM (2 of 3) [19/1/2003 1:29:30]

War Story: Covering Chessboards

attacked. The second pass filters the list by considering blocking and reports any boards with n or fewer safe squares. The advantage of separating weak and strong attack computations is that weak attack is faster to compute (no blocking to worry about), and yet the strong attack set is always a subset of the weak attack set. Whenever there was a non-weakly-threatened square, the position could be pruned. This program was efficient enough to complete the search on a machine as slow as a 1988-era IBM PCRT in under one day. More details of our searching procedure and results appear in our paper [RHS89]. It did not find a single position covering all 64 squares with the bishops on opposite colored squares. However, our program showed that it is possible to cover the board with seven pieces if a queen and a knight can occupy the same square. The take-home lesson of this war story should be clear. Clever pruning can make short work of surprisingly hard combinatorial search problems.

Next: Heuristic Methods Up: Combinatorial Search and Heuristic Previous: Bandwidth Minimization Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE90.HTM (3 of 3) [19/1/2003 1:29:30]

Heuristic Methods

Next: Simulated Annealing Up: Combinatorial Search and Heuristic Previous: War Story: Covering Chessboards

Heuristic Methods The techniques we have discussed thus far seek to find the optimal answer to a combinatorial problem as quickly as possible. Traditional algorithmic methods fail whenever the problem is provably hard (as discussed in Chapter

), or the problem is not clean enough to lead to a nice formulation.

Heuristic methods provide a way to approach difficult combinatorial optimization problems. Combinatorial search gives us a method to construct possible solutions and find the best one, given a function that measures how good each candidate solution is. However, there may be no algorithm to find the best solution short of searching all configurations. Heuristic methods such as simulated annealing, genetic algorithms, and neural networks provide general ways to search for good but not optimal solutions. In this section we discuss such heuristic methods. Each of these three techniques relies on a simple model of a real-world physical process. We devote the bulk of our attention to simulated annealing, which is the easiest method to apply in practice, as well as the most reliable.

●

● ●

Simulated Annealing ❍ Traveling Salesman Problem ❍ Maximum Cut ❍ Independent Set ❍ Circuit Board Placement Neural Networks Genetic Algorithms

Algorithms file:///E|/BOOK/BOOK2/NODE91.HTM (1 of 2) [19/1/2003 1:29:30]

Heuristic Methods

Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE91.HTM (2 of 2) [19/1/2003 1:29:30]

Simulated Annealing

Next: Traveling Salesman Problem Up: Heuristic Methods Previous: Heuristic Methods

Simulated Annealing

The inspiration for simulated annealing comes from the physical process of cooling molten materials down to the solid state. When molten steel is cooled too quickly, cracks and bubbles form, marring its surface and structural integrity. To end up with the best final product, the steel must be cooled slowly and evenly. Annealing is a metallurgical technique that uses a disciplined cooling schedule to efficiently bring the steel to a low-energy, optimal state. In thermodynamic theory, the energy state of a system is described by the energy state of each of the particles constituting it. The energy state of each particle jumps about randomly, with such transitions governed by the temperature of the system. In particular, the probability of transition from energy to at temperature T is given by

where

is a constant, called Boltzmann's constant.

What does this formula mean? Consider the value of the exponent under different conditions. The probability of moving from a high-energy state to a lower-energy state is very high. However, there is also a non-zero probability of accepting a transition into a high-energy state, with small energy jumps much more likely than big ones. The higher the temperature, the more likely such energy jumps will occur. What relevance does this have for combinatorial optimization? A physical system, as it cools, seeks to go to a minimumenergy state. For any discrete set of particles, minimizing the total energy is a combinatorial optimization problem. Through random transitions generated according to the above probability distribution, we can simulate the physics to solve arbitrary combinatorial optimization problems.

Simulated-Annealing()

Create initial solution S

Initialize temperature t

repeat

for i=1 to iteration-length do

Generate a random transition from S to

file:///E|/BOOK/BOOK2/NODE92.HTM (1 of 3) [19/1/2003 1:29:34]

Simulated Annealing

If

then

else if then

Reduce temperature t

until (no change in C(S))

Return S

There are three components to any simulated annealing algorithm for combinatorial search: ●

●

●

Concise problem representation - The problem representation includes both a representation of the solution space and an appropriate and easily computable cost function C() measuring the quality of a given solution. Transition mechanism between solutions - To move from one state to the next, we need a collection of simple transition mechanisms that slightly modify the current solution. Typical transition mechanisms include swapping the position of a pair of items or inserting/deleting a single item. Ideally, the effect that these incremental changes have on measuring the quality of the solution can be computed incrementally, so cost function evaluation takes time proportional to the size of the change (typically constant) instead of linear in the size of the solution. Cooling schedule - These parameters govern how likely we are to accept a bad transition as a function of time. At the beginning of the search, we are eager to use randomness to explore the search space widely, so the probability of accepting a negative transition is high. As the search progresses, we seek to limit transitions to local improvements and optimizations. The cooling schedule can be regulated by the following parameters: ❍ Initial system temperature - Typically . ❍ Temperature decrement function - Typically , where . This implies an exponential decay in the temperature, as opposed to a linear decay. ❍ Number of iterations between temperature change - Typically, 100 to 1,000 iterations might be permitted before lowering the temperature. ❍ Acceptance criteria - A typical criterion is to accept any transition from to when and to accept a negative transition whenever

❍

where r is a random number . The constant c normalizes this cost function, so that almost all transitions are accepted at the starting temperature. Stop criteria - Typically, when the value of the current solution has not changed or improved within the last iteration or so, the search is terminated and the current solution reported.

Creating the proper cooling schedule is somewhat of a trial and error process. It might pay to start from an existing implementation of simulated annealing, pointers to which are provided in Section

file:///E|/BOOK/BOOK2/NODE92.HTM (2 of 3) [19/1/2003 1:29:34]

.

Simulated Annealing

We provide several examples below to demonstrate how these components can lead to elegant simulated annealing algorithms for real combinatorial search problems.

● ● ● ●

Traveling Salesman Problem Maximum Cut Independent Set Circuit Board Placement

Next: Traveling Salesman Problem Up: Heuristic Methods Previous: Heuristic Methods Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE92.HTM (3 of 3) [19/1/2003 1:29:34]

Traveling Salesman Problem

Next: Maximum Cut Up: Simulated Annealing Previous: Simulated Annealing

Traveling Salesman Problem The solution space for traveling salesman consists of the set of all (n-1)! possible circular permutations of the vertices. A candidate solution can thus be represented using an array S of n-1 vertices, where defines the (i+1)st vertex on the tour starting from . The cost function evaluating a candidate solution is equally straightforward, for we can sum up the costs of the edges defined by S.

Figure: Improving a TSP tour by swapping a pair of edges The most obvious transition mechanism would be to swap the current tour positions of a random pair of vertices and . This changes up to eight edges on the tour, deleting the edges currently adjacent to both and , and adding their replacements. Better would be to swap two edges on the tour with two others that replace it, as shown in Figure . Since only four edges change in the tour, the transitions can be performed and evaluated faster. Faster transitions mean that we can evaluate more positions in the given amount of time. In practice, problem-specific heuristics for TSP outperform simulated annealing, but the simulated annealing solution works admirably, considering it uses very little knowledge about the problem.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE93.HTM [19/1/2003 1:29:34]

Maximum Cut

Next: Independent Set Up: Simulated Annealing Previous: Traveling Salesman Problem

Maximum Cut For a weighted graph G, the maximum cut problem seeks to partition the vertices into sets and so as to maximize the weight (or number) of edges with one vertex in each set. When the graph specifies an electronic circuit, the maximum cut in the graph defines the largest amount of data communication that can take place in the circuit simultaneously. As discussed in catalog Section complete version of graph partitioning.

, maximum cut is an NP-

How can we formulate maximum cut for simulated annealing? The solution space consists of all possible vertex partitions; we save a factor of two over all vertex subsets because we can assume that vertex is fixed to be on the left side of the partition. The subset of vertices accompanying it can be represented using a bit vector. The cost of a solution will be the sum of the weights cut in the current configuration. A natural transition mechanism is to select one vertex at random and move it across the partition by simply flipping the corresponding bit in the bit vector. The change in the cost function will be the weight of its old neighbors minus the weight of its new neighbors, so it can be computed in time proportional to the degree of the vertex. This kind of simple, natural modeling is the right type of heuristic to seek in practice.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE94.HTM [19/1/2003 1:29:35]

Independent Set

Next: Circuit Board Placement Up: Simulated Annealing Previous: Maximum Cut

Independent Set An independent set of a graph G is a subset of vertices S such that there is no edge with both endpoints in S. The maximum independent set of a graph is the largest such empty induced subgraph. The need to find large independent sets arises in dispersion problems associated with facility location and coding theory, as discussed in catalog Section

.

The natural state space for a simulated annealing solution would be all subsets of the vertices, represented as a bit vector. As with maximum cut above, a simple transition mechanism would be to add or delete one vertex from S. One natural cost function for subset S might be 0 if S contains an edge, and |S| if it is indeed an independent set. This function ensures that we work towards an independent set at all times. However, this condition is strict enough that we are liable to move only in a narrow portion of the possible search space. More flexibility in the search space and quicker cost function computations can result from allowing non-empty graphs at the early stages of cooling. Better in practice would be a cost function like , where is a constant, T is the temperature, and is the number of edges in the subgraph induced by S. The dependence of C(S) on T ensures that the search will drive the edges out faster as the system cools.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE95.HTM [19/1/2003 1:29:36]

Circuit Board Placement

Next: Neural Networks Up: Simulated Annealing Previous: Independent Set

Circuit Board Placement In designing printed circuit boards, we are faced with the problem of positioning modules (typically integrated circuits) on the board. Desired criteria in a layout include (1) minimizing the area or aspect ratio of the board, so that it properly fits within the allotted space, and (2) minimizing the total or longest wire length in connecting the components. Circuit board placement is an example of the kind of messy, multicriterion optimization problems for which simulated annealing is ideally suited. Formally, we are given a collection of a rectangular modules , each with associated dimensions . Further, for each pair of modules , we are given the number of wires that must connect the two modules. We seek a placement of the rectangles that minimizes area and wire-length, subject to the constraint that no two rectangles overlap each other. The state space for this problem must describe the positions of each rectangle. To provide a discrete representation, the rectangles can be restricted to lie on vertices of an integer grid. Reasonable transition mechanisms including moving one rectangle to a different location, or swapping the position of two rectangles. A natural cost function would be

where , , and are constants governing the impact of these components on the cost function. Presumably, should be an inverse function of temperature, so after gross placement it adjusts the rectangle positions so they are distinct. Simulated annealing performs well on such module placement problems. Indeed, a similar application appeared in the original paper on simulated annealing [KGV83]. More details on these and other applications appear in [AK89].

Algorithms Mon Jun 2 23:33:50 EDT 1997 file:///E|/BOOK/BOOK3/NODE96.HTM [19/1/2003 1:29:37]

Neural Networks

Next: Genetic Algorithms Up: Heuristic Methods Previous: Circuit Board Placement

Neural Networks

Neural networks are a computational paradigm inspired by the architecture of the human brain. The intuition is that since brains are good at solving problems, machines built in the same way should be, too.

The basic computational component of the brain is a neuron, a simple unit that produces a non-linear, weighted sum of its inputs, which are connections from other neurons. Neural networks are weighted digraphs with neurons as vertices and weights on edges denoting the connection strength of the pair. Brains are very good at learning and recognizing certain patterns. Learning in brains seems to work by adding connections between different pairs of neurons and changing the strengths of the connections. Modifying connection strength in response to training examples provides a natural way to ``teach'' a neural network. Although there have been attempts to apply neural networks to solving combinatorial optimization problems, the successes have been rather limited. Simulated annealing is a much more straightforward and efficient approach to optimization. Neural networks have been more successful in classification and forecasting applications, such as optical character recognition, gene prediction, and stock-market time-series prediction. A set of features for the given patterns is selected, and each training example is represented in terms of its features. The network is trained on a series of positive and negative examples, with the strengths of the connections adjusted to recognize these examples. Output cells for each class of item are provided and the strength of these cells on a given input used to determine the classification. Once the network is trained, feature vectors corresponding to unknown items can be entered and a classification made. Because neural networks are black boxes, with the strength of edges adjusted only by the training examples, there is usually no way to figure out exactly why they are making the decisions that they are. A particularly amusing instance where this led to trouble is reported in Section useful in certain pattern-recognition applications.

file:///E|/BOOK/BOOK3/NODE97.HTM (1 of 2) [19/1/2003 1:29:37]

. Still, they can be

Neural Networks

Next: Genetic Algorithms Up: Heuristic Methods Previous: Circuit Board Placement Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE97.HTM (2 of 2) [19/1/2003 1:29:37]

Genetic Algorithms

Next: War Story: Annealing Arrays Up: Heuristic Methods Previous: Neural Networks

Genetic Algorithms

Genetic algorithms draw their inspiration from evolution and natural selection. Through the process of natural selection, organisms adapt to optimize their chances for survival in a given environment. Random mutations occur to the genetic description of an organism, which is then passed on to its children. Should a mutation prove helpful, these children are more likely to survive to reproduce. Should it be harmful, these children are less likely to reproduce, so the bad trait will die with them. Genetic algorithms maintain a ``population'' of solution candidates for the given problem. Elements are drawn at random from this population and allowed to ``reproduce'', by combining some aspects of the two parent solutions. The probability that an element is chosen to reproduce is based on its ``fitness'', essentially a function of the cost of the solution it represents. Eventually, unfit elements die from the population, to be replaced by successful-solution offspring. The idea behind genetic algorithms is extremely appealing. However, they just don't seem to work as well on practical combinatorial optimization problems as simulated annealing does. There are two primary reasons for this. First, it is quite unnatural to model most applications in terms of genetic operators like mutation and crossover on bit strings. The pseudobiology adds another level of complexity between you and your problem. Second, genetic algorithms take a very long time on non-trivial problems. The crossover and mutation operations make no real use of problem-specific structure, so a large fraction of transitions lead to inferior solutions, and convergence is slow. Indeed, the analogy with evolution, where significant improvements require millions of years, can be quite appropriate. We will not discuss genetic algorithms further, in order to discourage you from considering them for your applications. However, pointers to implementations of genetic algorithms are provided in Section if you really insist on playing with them.

Next: War Story: Annealing Arrays Up: Heuristic Methods Previous: Neural Networks

file:///E|/BOOK/BOOK3/NODE98.HTM (1 of 2) [19/1/2003 1:29:38]

Genetic Algorithms

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE98.HTM (2 of 2) [19/1/2003 1:29:38]

War Story: Annealing Arrays

Next: Parallel Algorithms Up: Combinatorial Search and Heuristic Previous: Genetic Algorithms

War Story: Annealing Arrays reported how we used advanced data structures to simulate a new method The war story of Section for sequencing DNA. Our method, interactive sequencing by hybridization (SBH), involved building arrays of specific oligonucleotides on demand. Although the simulation results were very promising to us, most biologists we encountered were suspicious. They needed to see our technique proven in the lab before they would take it seriously. But we got lucky. A biochemist at Oxford University, got interested in our technique, and moreover he had in his laboratory the equipment we needed to test it out. The Southern Array Maker [Sou96], manufactured by Beckman Instruments, could prepare discrete oligonucleotide sequences in 64 parallel rows across a polypropylene substrate. The device constructs arrays by appending single characters to shows how to construct an array of all each cell along specific rows and columns of arrays. Figure purine (A or G) 4-mers by building the prefixes along rows and the suffixes along columns. This technology provided an ideal environment for testing the feasibility of interactive SBH in a laboratory, because with proper programming it gave an inexpensive way to fabricate a wide variety of oligonucleotide arrays on demand.

Figure: A prefix-suffix array of all purine 4-mers. But we had to provide the proper programming. Fabricating complicated arrays requires solving a difficult combinatorial problem. We were given as input a set S of n strings (representing oligonucleotides) to fabricate an array (where m=64 on the Southern apparatus). We had to produce a schedule of row and column commands to realize the set of strings S. We proved that the problem of designing dense arrays was NP-complete, but that didn't really matter. My student Ricky Bradley and I had to solve it anyway.

file:///E|/BOOK/BOOK3/NODE99.HTM (1 of 4) [19/1/2003 1:29:40]

War Story: Annealing Arrays

``If it's hard, it's hard. We are going to have to use a heuristic,'' I told him. ``So how do we model this problem?'' ``Well, for each string we can identify the possible prefix and suffix pairs that will realize it. For example, the string `ACC' can be realized in four different ways: prefix `' and suffix `ACC', prefix `A' and suffix `CC', prefix `AC' and suffix `C', or prefix `ACC' and suffix `'. We seek the smallest set of prefixes and suffixes that together realize all the given strings,'' Ricky said. ``Good. This gives us a natural representation for simulated annealing. The state space will consist of all possible subsets of prefixes and suffixes. The natural transitions between states might include inserting or deleting strings from our subsets, or swapping a pair in or out.'' ``What's a good cost function?'' he asked. ``Well, we need as small an array as possible that covers all the strings. How about something like the maximum of number of rows (prefixes) or columns (suffixes) used in our array, plus the number of strings from S that are not yet covered. Try it and let's see what happens.'' Ricky went off and implemented a simulated annealing program along these lines. Printing out the state of the solution each time a transition was accepted, it was fun to watch. Starting from a random solution, the program quickly kicked out unnecessary prefixes and suffixes, and the array began shrinking rapidly in size. But after several hundred iterations, progress started to slow. A transition would knock out an unnecessary suffix, wait a while, then add a different suffix back again. After a few thousand iterations, no real improvement was happening. ``The program doesn't seem to recognize when it is making progress. The evaluation function only gives credit for minimizing the larger of the two dimensions. Why not add a term to give some credit to the other dimension.'' Ricky changed the evaluation function, and we tried again. This time, the program did not hesitate to improve the shorter dimension. Indeed, our arrays started to be skinny rectangles instead of squares. ``OK. Let's add another term to the evaluation function to give it points for being roughly square.'' Ricky tried again. Now the arrays were the right shape, and progress was in the right direction. But the progress was slow. ``Too many of the prefix/suffix insertion moves don't really seem to affect many strings. Maybe we should skew the random selections so that the important prefix/suffixes get picked more often.'' Ricky tried again, Now it converged faster, but sometimes it still got stuck. We changed the cooling schedule. It did better, but was it doing well? Without a lower bound knowing how close we were to file:///E|/BOOK/BOOK3/NODE99.HTM (2 of 4) [19/1/2003 1:29:40]

War Story: Annealing Arrays

optimal, it couldn't really tell how good our solution was. We tweaked and tweaked until our program stopped improving. Our final solution refined the initial array by applying the following random moves: ● ● ● ● ● ●

swap - swap a prefix/suffix on the array with one that isn't. add - add a random prefix/suffix to the array. delete - delete a random prefix/suffix from the array. useful add - add the prefix/suffix with the highest usefulness to the array. useful delete - delete the prefix/suffix with the lowest usefulness from the array. string add - randomly select a string not on the array, and add the most useful prefix and/or suffix that covers this string (additional preference is given to a prefix/suffix whose corresponding suffix/prefix is already on the array).

A standard annealing schedule was used, with an exponentially decreasing temperature (dependent upon the problem size) and a temperature-dependent Boltzmann criterion for accepting states that have higher costs. Our final cost function was defined as

where max is the size of the maximum chip dimension, min is the size of the minimum chip dimension, , and is the number of strings in S currently on the chip. Careful analysis of successful moves over the course of the annealing process suggested a second phase of annealing to speed convergence. Once the temperature reaches a predetermined cutoff point, the temperature schedule was changed to force the temperature to decrease more rapidly, and the probability distribution for choosing moves was altered to include only swap, add, and delete, with preference given to swap moves. This modification sped up late convergence, which had been slower than it was in the early stages of the annealing process.

file:///E|/BOOK/BOOK3/NODE99.HTM (3 of 4) [19/1/2003 1:29:40]

War Story: Annealing Arrays

Figure: Compression of the HIV array by simulated annealing - after 0, 500, 1,000, and 5,750 iterations How well did we do? As reported in our paper [BS97], Figure

shows the convergence of a custom

array consisting of the 5,716 unique 7-mers of the HIV-virus. Figure shows snapshots of the state of the chip at four points during the annealing process (0, 500, 1,000, and the final chip at 5,750 iterations). Black pixels represent the first occurrence of an HIV 7-mer, while white pixels represent either duplicated HIV 7-mers or strings not in the HIV input set. The final chip size here is , quite an improvement over the initial size of . It took about fifteen minutes' worth of computation on a desktop workstation to complete the optimization, which was perfectly acceptable for the application. But how well did we do? Since simulated annealing is only a heuristic, we really don't know how close to optimal our solution is. I think we did pretty well, but I can't really be sure. In conclusion, simulated annealing can be the right way to handle complex optimization problems. However, to get the best results, expect to spend more time tweaking and refining your program than you did in writing it in the first place. This is dirty work, but sometimes you have to do it.

Next: Parallel Algorithms Up: Combinatorial Search and Heuristic Previous: Genetic Algorithms Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE99.HTM (4 of 4) [19/1/2003 1:29:40]

Parallel Algorithms

Next: War Story: Going Nowhere Up: Combinatorial Search and Heuristic Previous: War Story: Annealing Arrays

Parallel Algorithms Two heads are better than one, and more generally, n heads are better than n-1. In our era of computing plenty, parallel processing seems like an exciting technique for solving combinatorial optimization problems. Today many facilities contain networks of workstations, most of them idle at night and underutilized during the day. Why not put them to work? Parallelism seems like the easy way out of hard problems. Indeed, sometimes, for some problems, parallel algorithms are the most effective solution. High-resolution, real-time graphics applications must render thirty frames per second for realistic animation. Assigning each frame to a distinct processor, or dividing each image into regions assigned to different processors might be the only way to get the job done in time. Large systems of linear equations for scientific applications are routinely solved in parallel. However, there are several pitfalls associated with parallel algorithms that one should be aware of: ●

●

There is often a small upper bound on the potential win - Suppose that you have access to twenty workstations that can be devoted exclusively to your job. Potentially, these could be used to speed up the fastest sequential program by up to a factor of twenty. That is nice, but much greater performance gains are potentially possible by finding a better sequential algorithm and implementing that. Your time spent parallelizing a code might well be better spent enhancing the sequential version. Performance-tuning tools such as profilers are better developed for sequential machines than for parallel models. Speedup means nothing - Suppose my parallel program runs 16 times faster on a 16-processor machine then it does on one processor. That's great, isn't it? If you always get linear speedup and have an arbitrary number of processors, you will eventually beat any sequential algorithm. However, a carefully designed sequential algorithm can often beat an easily parallelized code running on a typical parallel machine. The one-processor parallel version of your algorithm is likely to be a crummy sequential algorithm, so measuring speedup typically provides an unfair test of the benefits of parallelism. The classic example of this occurs in the minimax game-tree search algorithms used in computer chess programs. Brute-force tree search is embarrassingly easy to parallelize; just put each subtree on a different processor. However, a lot of work gets wasted because the same positions get considered on different machines. Moving from brute-force search to the more clever alpha-beta

file:///E|/BOOK/BOOK/NODE100.HTM (1 of 2) [19/1/2003 1:29:41]

Parallel Algorithms

●

pruning algorithm can easily save over 99.99% of the work, thus dwarfing any benefits of parallel brute-force search. Alpha-beta can be parallelized, but not easily, and speedups are typically limited to a factor of six or so regardless of how many processors you have. Parallel algorithms are tough to debug - Unless your problem can be decomposed into several independent jobs, the different processors will have to communicate with each other in order to end up with the correct final result. Unfortunately, the non-deterministic nature of this communication makes parallel programs notoriously difficult to debug. Perhaps the best example is Deep Blue, the world-champion chess computer. Although it beat Kasparov, over the years it has lost several games in embarrassing fashion due to bugs, mostly associated with its extensive parallelism.

I recommend considering parallel processing only after repeated attempts at solving the problem sequentially prove too slow. Even then, I would restrict attention to algorithms that parallelize the problem by partitioning the input into distinct tasks, where no communication is needed between the processors, except in collecting the final results. Such large-grain, naive parallelism can be simple enough to be readily implementable and debuggable, because it really reduces to producing a good sequential implementation. Still, there can be pitfalls in this approach, as discussed in the war story below.

Next: War Story: Going Nowhere Up: Combinatorial Search and Heuristic Previous: War Story: Annealing Arrays Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE100.HTM (2 of 2) [19/1/2003 1:29:41]

War Story: Going Nowhere Fast

Next: Exercises Up: Combinatorial Search and Heuristic Previous: Parallel Algorithms

War Story: Going Nowhere Fast , I related our efforts to build a fast program to test Waring's conjecture for pyramidal In Section numbers. At that point, my code was fast enough that it could complete the job in a few weeks running in the background on a desktop workstation. This option did not appeal to my supercomputing colleague, however. ``Why don't we do it in parallel?'' he suggested. ``After all, you have an outer loop doing the same type of calculation on each integer from 1 to 1,000,000,000. I can split this range of numbers into different intervals and run each one of these on a different processor. Watch, it will be easy.'' He set to work trying to do our computations on an Intel IPSC-860 hypercube using 32 nodes, with 16 megabytes of memory per node. However, instead of getting answers, over the next few weeks I was treated to a regular stream of e-mail about system reliability: ● ● ●

``Our code is running fine, except one processor died last night. I will rerun.'' ``This time the machine was rebooted by accident, so our long-standing job was killed.'' ``We have another problem. The policy on using our machine is that nobody can command the entire machine for more than thirteen hours, under any condition.''

Still, eventually, he rose to the challenge. Waiting until the machine was stable, he locked out 16 processors (half the computer), divided the integers from 1 to 1,000,000,000 into 16 equal-sized intervals, and ran each interval on its own processor. He spent the next day fending off angry users who couldn't get their work done because of our rogue job. The instant the first processor completed analyzing the numbers from 1 to 62,500,000, he announced to all the people yelling at him that the other processors would soon follow. But they didn't. He failed to realize that the time to test each integer increased as the numbers got larger. After all, it would take longer to test whether 1,000,000,000 could be expressed as the sum of three pyramidal number than it would for 100. Thus at slower and slower intervals, each new processor would announce its completion. Because of the architecture of the hypercube, he couldn't return any of the processors until our entire job was completed. Eventually, half the machine and most of its users were held hostage by one, final interval. file:///E|/BOOK/BOOK3/NODE101.HTM (1 of 2) [19/1/2003 1:29:41]

War Story: Going Nowhere Fast

When the job finally completed, the numbers were passed on to the Nobel Prize winner who had requested them. It turns out he had been curious about the problem because his father had made the conjecture back in 1928. There had never been a more important scientific reason to justify the computation in the first place. Indeed, no results from the computation ever appeared in print. What conclusions can be drawn from this? Before devoting heroic efforts to solve a problem efficiently, make sure that it really needs to be solved, and solved quickly. If you are going to parallelize a problem, be sure to balance the load carefully among the processors. Proper load balancing, using either back-of, would have significantly reduced the the-envelope calculations or the partition algorithm of Section time we needed the machine, and his exposure to the wrath of his colleagues.

Next: Exercises Up: Combinatorial Search and Heuristic Previous: Parallel Algorithms Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE101.HTM (2 of 2) [19/1/2003 1:29:41]

Exercises

Next: Implementation Challenges Up: Combinatorial Search and Heuristic Previous: War Story: Going Nowhere

Exercises 1. (*) A derangement is a permutation p of such that no item is in its proper position, i.e. for all . Write an efficient backtracking program with pruning that constructs all the derangements of n items. 2. (*) Multisets are allowed to have repeated elements. A multiset of n items may thus have fewer has only six different permutations: than n! distinct permutations. For example, , , , , , and . Design and implement an efficient algorithm for constructing all permutations of a multiset. 3. (*) Design and implement an algorithm for testing whether two graphs are isomorphic to each . With proper pruning, graphs other. The graph isomorphism problem is discussed in Section on hundreds of vertices can be tested reliably. 4. (**) Design and implement an algorithm for solving the subgraph isomorphism problem. Given graphs G and H, does there exist a subgraph H' of H such that G is isomorphic to H'. How does your program perform on such special cases of subgraph isomorphism as Hamiltonian cycle, clique, independent set, and graph isomorphism. 5. (*) Design and implement an algorithm for solving the set cover problem, discussed in Section . Use it to solve special-case vertex cover problems as well as general set cover problems. 6. (**) In the turnpike reconstruction problem, you are given n(n-1)/2 distances in sorted order. The problem is to find the positions of the points on the line that give rise to these distances. For can be determined by placing the second point 1 unit from example, the distances the first, the third point 3 from the second, and the fourth point 2 from the third. Design and implement an efficient algorithm to report all solutions to the turnpike reconstruction problem. Exploit additive constraints when possible to minimize search. With proper pruning, problems with hundreds of points can be solved reliably.

file:///E|/BOOK/BOOK3/NODE102.HTM (1 of 2) [19/1/2003 1:29:42]

Exercises

Next: Implementation Challenges Up: Combinatorial Search and Heuristic Previous: War Story: Going Nowhere Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE102.HTM (2 of 2) [19/1/2003 1:29:42]

Problems and Reductions

Next: Simple Reductions Up: Intractable Problems and Approximations Previous: Intractable Problems and Approximations

Problems and Reductions Throughout this book we have encountered problems, such as the traveling salesman problem, for which we couldn't find any efficient algorithm. By the early 1970s, literally hundreds of problems were stuck in this swamp. The theory of NPcompleteness provided the tools needed to show that all of these problems were really the same thing. The key idea behind demonstrating the hardness of a problem is that of a reduction. Suppose that I gave you the following algorithm to solve the Bandersnatch problem:

Bandersnatch(G)

Translate the input G to an instance of the Bo-billy problem Y.

Call the subroutine Bo-billy on Y to solve this instance.

Return the answer of Bo-billy(Y) as the answer to Bandersnatch(G).

It is important to see that this algorithm correctly solves the Bandersnatch problem provided that the translation to Bo-billy always preserves the correctness of the answer. In other words, the translation has the property that for any instance of G, Bandersnatch(G) = Bo-billy(Y). A translation of instances from one type of problem to instances of another type such that the answers are preserved is called a reduction. Now suppose this reduction translates G to Y in O(P(n)) time. There are two possible implications: ●

●

If my Bo-billy subroutine ran in O(P'(n)), this means I could solve the Bandersnatch problem in O(P(n)+P'(n)) by spending the time to translate the problem and then the time to execute the Bo-Billy subroutine. If I know that is a lower bound on computing Bandersnatch, meaning there definitely exists no faster way to solve it, then must be a lower bound to compute Bo-billy. Why? If I could solve Bo-billy any faster, then I could solve Bandersnatch in faster time by using the above simulation, thus violating my lower bound. This implies that there can be no way to solve Bo-billy any faster than claimed.

This second argument is the approach that we will use to prove problems hard. Essentially, this reduction shows that Bobilly is at least as hard as Bandersnatch, and therefore once we believe that Bandersnatch is hard, we have a tool for proving other problems hard. Reductions, then, are operations that convert one problem into another. To describe them, we must be somewhat rigorous in our definition of a problem. A problem is a general question, with parameters for the input and conditions on what constitutes a satisfactory answer or solution. An instance is a problem with the input parameters specified. The difference file:///E|/BOOK/BOOK3/NODE105.HTM (1 of 2) [19/1/2003 1:29:43]

Problems and Reductions

can be made clear by an example. The traveling salesman problem is defined thus: Input: A weighted graph G. Output: Which tour

minimizes

?

Thus any weighted graph defines an instance of TSP. Each particular instance has at least one minimum cost tour. The general traveling salesman problem asks for an algorithm to find the optimal tour for all possible instances. Any problem with answers restricted to yes and no is called a decision problem. Most interesting optimization problems can be phrased as decision problems that capture the essence of the computation. For example, the traveling salesman decision problem can be defined thus: Input: A weighted graph G and integer k. ? It should be clear that the decision version captures the heart of the Output: Does there exist a TSP tour with cost traveling salesman problem, for if you had a program that gave fast solutions to the decision problem, you could do a binary search with different values of k to quickly hone in on the correct solution. Therefore, from now on we will talk only about decision problems, because it is easier to reduce one problem to another when the only possible answers to both are true or false.

Next: Simple Reductions Up: Intractable Problems and Approximations Previous: Intractable Problems and Approximations Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE105.HTM (2 of 2) [19/1/2003 1:29:44]

Simple Reductions

Next: Hamiltonian Cycles Up: Intractable Problems and Approximations Previous: Problems and Reductions

Simple Reductions Since they can be used either to prove hardness or to give efficient algorithms, reductions are powerful tools for the algorithm designer to be familiar with. The best way to understand reductions is to look at some simple ones.

● ● ●

Hamiltonian Cycles Independent Set and Vertex Cover Clique and Independent Set

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE106.HTM [19/1/2003 1:29:44]

Hamiltonian Cycles

Next: Independent Set and Vertex Up: Simple Reductions Previous: Simple Reductions

Hamiltonian Cycles The Hamiltonian cycle problem is one of the most famous in graph theory. It seeks a tour that visits each vertex of a given graph exactly once. It has a long history and many applications, as discussed in Section

. Formally, it is defined:

Input: An unweighted graph G. Output: Does there exist a simple tour that visits each vertex of G without repetition? Hamiltonian cycle has some obvious similarity to the traveling salesman problem. Both problems ask for a tour to visit each vertex exactly once. There are also differences between the two problems. TSP works on weighted graphs, while Hamiltonian cycle works on unweighted graphs. The following reduction from Hamiltonian cycle to traveling salesman shows that the similarities are greater than the differences:

HamiltonianCycle(G=(V,E))

Construct a complete weighted graph G'=(V',E') where V'=V.

n = |V|

for i = 1 to n do

for j = 1 to n do if

then w(i,j) = 1

else w(i,j) = 2

Return the answer to Traveling-Salesman(G',n).

The actual reduction is quite simple, with the translation from unweighted to weighted graph easily performed in linear time. Further, this translation is designed to ensure that the answers of the two problems will be identical. If the graph G , then this exact same tour will correspond to n edges in E', each with weight 1. has a Hamiltonian cycle Therefore, this gives a TSP tour of G' of weight exactly n. If G does not have a Hamiltonian cycle, then there can be no such TSP tour in G', because the only way to get a tour of cost n in G would be to use only edges of weight 1, which implies a Hamiltonian cycle in G. This reduction is both efficient and truth preserving. A fast algorithm for TSP would imply a fast algorithm for Hamiltonian cycle, while a hardness proof for Hamiltonian cycle would imply that TSP is hard. Since the latter is the case, this reduction shows that TSP is hard, at least as hard as Hamiltonian cycle. file:///E|/BOOK/BOOK3/NODE107.HTM (1 of 2) [19/1/2003 1:29:45]

Hamiltonian Cycles

Next: Independent Set and Vertex Up: Simple Reductions Previous: Simple Reductions Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE107.HTM (2 of 2) [19/1/2003 1:29:45]

Independent Set and Vertex Cover

Next: Clique and Independent Set Up: Simple Reductions Previous: Hamiltonian Cycles

Independent Set and Vertex Cover

The vertex cover problem, discussed more thoroughly in Section contacts each edge in a graph. More formally: Input: A graph G=(V, E) and integer

, asks for a small set of vertices that

.

Output: Is there a subset S of at most k vertices such that every

has at least one vertex in S?

Figure: Circled vertices form a vertex cover, the dark vertices an independent set It is trivial to find a vertex cover of a graph, for the cover can consist of all of the vertices. More tricky is to cover the edges using as small a set of vertices as possible. For the graph of Figure eight vertices are sufficient to cover. A set of vertices S of graph G is independent if there are no edges (x,y) where

and

, four of the

, meaning

there are no edges between any two vertices in the independent set. As discussed in Section , the independent set problem arises in facility location problems. The maximum independent set decision problem is formally defined: Input: A graph G and integer

.

Output: Does there exist an independent set of k vertices in G? Both vertex cover and independent set are file:///E|/BOOK/BOOK3/NODE108.HTM (1 of 2) [19/1/2003 1:29:46]

Independent Set and Vertex Cover

problems that revolve around finding special subsets of vertices, the first with representatives of every edge, the second with no edges. If S is the vertex cover of G, the remaining vertices S-V must form an independent set, for if there were an edge with both vertices in S-V, then S could not have been a vertex cover. This gives us a reduction between the two problems:

VertexCover(G,k)

G' = G

k' = |V| - k

Return the answer to IndependentSet(G',k')

Again, a simple reduction shows that the problems are identical. Notice how this translation occurs without any knowledge of the answer. We transform the input, not the solution. This reduction shows that the hardness of vertex cover imples that independent set must also be hard. It is easy to reverse the roles of the two problems in this reduction, thus proving that both of these problems are equally hard.

Next: Clique and Independent Set Up: Simple Reductions Previous: Hamiltonian Cycles Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE108.HTM (2 of 2) [19/1/2003 1:29:46]

Clique and Independent Set

Next: Satisfiability Up: Simple Reductions Previous: Independent Set and Vertex

Clique and Independent Set

Consider the clique problem, further discussed in Section

:

Figure: A small graph with a five-vertex clique Input: A graph G=(V,E) and integer

.

Output: Does the graph contain a clique of j vertices; i.e. is there a subset

, where

, such that every pair

of vertices in S defines an edge of G? For example, the graph in Figure contains a clique of five vertices. In the independent set problem, we looked for a subset S with no edges between two vertices of S. However, for a clique, we insist that there always be an edge between two vertices. A reduction between these problems results by reversing the roles of edges and non-edges, an operation known as complementing the graph:

IndependentSet(G,k)

Construct a graph G=(V',E') where V'=V, and

For all (i,j) not in E, add (i,j) to E'

Return the answer to Clique(G',k)

file:///E|/BOOK/BOOK3/NODE109.HTM (1 of 2) [19/1/2003 1:29:47]

Clique and Independent Set

These last two reductions provide a chain linking three different problems. The hardness of clique is implied by the hardness of independent set, which is implied by the hardness of vertex cover. By constructing reductions in a chain, we link together pairs of problems in implications of hardness. Our work is done as soon as all these chains begin with a single problem that is accepted as hard. Satisfiability is the problem that serves as the first link in this chain.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE109.HTM (2 of 2) [19/1/2003 1:29:47]

Satisfiability

Next: The Theory of NP-Completeness Up: Intractable Problems and Approximations Previous: Clique and Independent Set

Satisfiability To prove the hardness of different problems using reductions, we need to start with a single problem that is absolutely, certifiably, undeniably hard. The mother of all NP-complete problems is a logic problem named satisfiability: Input: A set of Boolean variables V and a set of clauses C over V. Output: Does there exist a satisfying truth assignment for C, i.e. a way to set the variables either true or false so that each clause contains at least one true literal? This can be made clearer with two over the Boolean variables . We use to examples. Suppose that denote the complement of the variable , so we would get credit for satisfying a particular clause , or a clause containing if . Therefore, satisfying a particular set of containing if clauses involves making a series of n true or false decisions, trying to find the right truth assignment to satisfy all of them. This example set of clauses can be satisfied by simply setting or . However, consider the set of clauses . There can be no in order to satisfy the third clause, which means that satisfying assignment because must be to satisfy the second clause, which then leaves the first clause unsatisfiable. Although you must be try, and you try, and you try and you try, you can't get no satisfaction.

● ●

The Theory of NP-Completeness 3-Satisfiability

Algorithms file:///E|/BOOK/BOOK3/NODE110.HTM (1 of 2) [19/1/2003 1:29:48]

Satisfiability

Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE110.HTM (2 of 2) [19/1/2003 1:29:48]

The Theory of NP-Completeness

Next: 3-Satisfiability Up: Satisfiability Previous: Satisfiability

The Theory of NP-Completeness For a variety of social and technical reasons, it is well accepted that satisfiability is a hard problem, one for which no worst-case polynomial-time algorithm exists. Literally every top-notch algorithm expert in the world (and countless lesser lights) has directly or indirectly tried to come up with a fast algorithm to test whether a given set of clauses is satisfiable, but all have failed. Further, many strange and impossibleto-believe things in the field of computational complexity have been shown to be true if there exists a fast satisfiability algorithm. Satisfiability is a hard problem, and it is important to accept this. See Section for more on the satisfiability problem and its applications. The theory of NP-completeness rests on a foundation of rigorous but subtle definitions from automata and formal language theory. This terminology is typically confusing to or misused by beginners who lack a mastery of these foundations, and it is not really essential to the practical aspects of designing and applying reductions. For completeness, however, we briefly define the key terms below. A problem is said to be polynomial (or in the class P) if it can be solved in time polynomial in its size. A problem is said to be nondeterministically polynomial (or in the class NP) if a conjectured answer can be verified in time polynomial in its size. The traveling salesman decision problem is not known to be in P, because there is no known polynomial-time algorithm for it. However, the problem is in NP, because if we are given a candidate tour, we can efficiently add up the cost of the associated edges and verify whether the total is at most the cost bound k. It is typically straightforward to verify whether the answer to a problem is correct, and it certainly can be no harder than actually finding the answer in the first place.

Through a complicated proof, it has been established that satisfiability is at least as hard as any problem in NP. This means that if a fast (i.e. polynomial-time) algorithm is discovered for solving satisfiability, this will yield a fast algorithm for every problem in NP. Since essentially every problem mentioned this book is in NP, this would be an enormously powerful and surprising result. We say that a problem is NPhard if, like satisfiability, it is at least as hard as any problem in NP. We say that a problem is NPcomplete if it is NP-hard, and also in NP itself. Because NP is such a large class of problems, most NPhard problems you encounter will in fact be complete, and the issue can always be settled by giving a (usually simple) verification strategy for the problem.

file:///E|/BOOK/BOOK3/NODE111.HTM (1 of 2) [19/1/2003 1:29:48]

The Theory of NP-Completeness

Next: 3-Satisfiability Up: Satisfiability Previous: Satisfiability Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE111.HTM (2 of 2) [19/1/2003 1:29:48]

3-Satisfiability

Next: Difficult Reductions Up: Satisfiability Previous: The Theory of NP-Completeness

3-Satisfiability Satisfiability's role as the first NP-complete problem implies that the problem is hard to solve in the worst case, but certain instances of the problem are not necessarily so tough. Suppose that each clause contains exactly one literal. To satisfy such a clause, we have to appropriately set that literal, so we can repeat this argument for every clause in the problem instance. Only when we have two clauses that directly contradict each other, such as , will the set not be satisfiable. Since clause sets with only one literal per clause are easy to satisfy, we are interested in slightly larger classes. Exactly what is the clause size at which the problem turns from polynomial to hard? This transition occurs when each clause contains three literals, the so-called 3-satisfiability problem, or 3SAT: Input: A collection of clauses C where each clause contains exactly 3 literals, over a set of Boolean variables V. Output: Is there a truth assignment to V such that each clause is satisfied? Since this is a more restricted problem than satisfiablity, the hardness of 3-SAT implies that satisfiability is hard. The converse isn't true, as the hardness of general satisfiability might depend upon having long clauses. We can show the hardness of 3-SAT using a reduction that translates every instance of satisfiability into an instance of 3SAT without changing the result of whether it is satisfiable. This reduction transforms each clause independently based on its length, by adding new Boolean variables along the way. Suppose clause contained k literals: ●

●

● ●

If k=1, meaning that , we create two new variables and four new 3-literal clauses: , , , . Note that the only way that all four of these clauses , which must be the case if the original were to be can be simultaneously satisfied is if satisfied. If k=2, meaning that , we create one new variable and two new clauses: , . Again, the only way to satisfy both of these clauses is to have at least one of and be true. If k=3, meaning that , we copy into the 3-SAT instance unchanged: . If k>3, meaning that , create n-3 new variables and n-2 new clauses in a chain,

file:///E|/BOOK/BOOK3/NODE112.HTM (1 of 2) [19/1/2003 1:29:50]

3-Satisfiability

where for

,

,

, and

.

The most complicated case is that of the large clauses. If none of the original variables are , then there are not enough additional variables to be able to satisfy all of the new subclauses. You can satisfy by setting , but this forces , and so on until finally cannot be satisfied. But , then we have n-3 free variables and n-3 remaining 3-clauses, so we can if any single literal satisfy each of them. This transform takes O(m+n) time if there were n clauses and m total literals in the SAT instance. Since any SAT solution also satisfies the 3-SAT instance and any 3-SAT solution sets the variables giving a SAT solution, the transformed problem is equivallent to the original. Note that a slight modification to this construction would serve to prove that 4-SAT, 5-SAT, or any -SAT is also NP-complete. However, this construction breaks down when we try to use it for 2SAT, since there is no way to stuff anything into the chain of clauses. It turns out that resolution gives a polynomial-time algorithm for 2-SAT, as discussed in Section

.

Next: Difficult Reductions Up: Satisfiability Previous: The Theory of NP-Completeness Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE112.HTM (2 of 2) [19/1/2003 1:29:50]

Difficult Reductions

Next: Integer Programming Up: Intractable Problems and Approximations Previous: 3-Satisfiability

Difficult Reductions Now that both satisfiability and 3-SAT are known to be hard, we can use either of them in reductions. What follows are a pair of more complicated reductions, designed to serve both as examples for how to proceed and to increase our repertoire of known hard problems from which we can start. Many reductions are quite intricate, because we are essentially programming one problem in the language of a significantly different problem. One perpetual point of confusion is getting the direction of the reduction right. Recall that we must transform every instance of a known NP-complete problem into an instance of the problem we are interested in. If we perform the reduction the other way, all we get is a slow way to solve the problem of interest, by using a subroutine that takes exponential time. This always is confusing at first, for this direction of reduction seems bass-ackwards. Check to make sure you understand the direction of reduction now, and think back to this whenever you get confused.

● ●

Integer Programming Vertex Cover

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE113.HTM [19/1/2003 1:29:50]

Integer Programming

Next: Vertex Cover Up: Difficult Reductions Previous: Difficult Reductions

Integer Programming As discussed in Section , integer programming is a fundamental combinatorial optimization problem. It is best thought of as linear programming with the variables restricted to take only integer (instead of real) values. Input: A set V of integer variables, a set of inequalities over V, a maximization function f(V), and an integer B. Output: Does there exist an assignment of integers to V such that all inequalities are true and Consider the following two examples. Suppose

A solution to this would be following problem:

the maximum value of f(v) is associated decision problem.

,

?

. Not all problems have realizable solutions, however. For the

(given the constraints), and so there can be no solution to the

We show that integer programming is hard using a reduction from 3-SAT. For this particular reduction, general satisfiability would work just as well, although usually 3-SAT makes reductions easier. In which direction must the reduction go? We want to prove integer programming that is hard, and we know that 3-SAT is hard. If I could solve 3-SAT using integer programming and integer programming file:///E|/BOOK/BOOK3/NODE114.HTM (1 of 3) [19/1/2003 1:29:52]

Integer Programming

were easy, this would mean that satisfiability would be easy. Now the direction should be clear; we have to translate 3-SAT into integer programming. What should the translation look like? Every satisfiability instance contains Boolean (true/false) variables and clauses. Every integer programming instance contains integer variables (values restricted to 0,1,2,...) and constraints. A reasonable idea is to make the integer variables correspond to Boolean variables and have constraints serve the same role as the clauses do in the original problem. Our translated integer programming problem will have twice as many variables as the SAT instance, one for each variable and one for its complement. For each variable in the set problem, we will add the following constraints: ●

To restrict each integer programming variable

to values of 0 or 1, we add constraints

●

and . Thus they correspond to values of and . To ensure that exactly one of the two integer programming variables associated with a given SAT variable is , add constraints so that .

in the 3-SAT instance, construct a constraint: . To satisfy For each clause this constraint, at least one the literals per clause must be set to 1, thus corresponding to a true literal. Satisfying this constraint is therefore equivalent to satisfying the clause. The maximization function and bound prove relatively unimportant, since we have already encoded the entire 3-SAT instance. By using and B=0, we ensure that they will not interfere with any variable assignment satisfying all the inequalities. Clearly, this reduction can be done in polynomial time. To establish that this reduction preserves the answer, we must verify two things: ●

●

Any SAT solution gives a solution to the IP problem - In any SAT solution, a literal corresponds to a 1 in the integer program, since the clause is satisfied. Therefore, the sum in each clause inequality is . Any IP solution gives a SAT solution - In any solution to this integer programming instance, all , then set literal . If , then set literal variables must be set to either 0 or 1. If . No Boolean variable and its complement can both be true, so it is a legal assignment, which must also satisfy all the clauses.

The reduction works both ways, so integer programming must be hard. Notice the following properties, which hold true in general for NP-complete: ●

●

The reduction preserved the structure of the problem. It did not solve the problem, just put it into a different format. The possible IP instances that can result from this transformation are only a small subset of all

file:///E|/BOOK/BOOK3/NODE114.HTM (2 of 3) [19/1/2003 1:29:52]

Integer Programming

●

possible IP instances. However, since some of them are hard, the general problem must be hard. The transformation captures the essence of why IP is hard. It has nothing to do with having big coefficients or big ranges on variables; since restricting them to 0/1 is enough. It has nothing to do with having inequalties with large numbers of variables. Integer programming is hard because satisfying a set of constraints is hard. A careful study of the properties needed for a reduction can tell us a lot about the problem.

Next: Vertex Cover Up: Difficult Reductions Previous: Difficult Reductions Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE114.HTM (3 of 3) [19/1/2003 1:29:52]

Vertex Cover

Next: Other NP-Complete Problems Up: Difficult Reductions Previous: Integer Programming

Vertex Cover Algorithmic graph theory proves to be a fertile ground for hard problems. The prototypical NP-complete graph problem is vertex cover, previously defined in Section Input: A graph G=(V, E) and integer

as follows:

.

Output: Is there a subset S of at most k vertices such that every has at least one vertex in S? Demonstrating the hardness of vertex cover proves more difficult than the previous reductions we have seen, because the structure of the two relevant problems is very different. A reduction from 3satisfiability to vertex cover has to construct a graph G and bound k from the variables and clauses of the satisfiability instance. First, we translate the variables of the 3-SAT problem. For each Boolean variable , we create two vertices and connected by an edge. To cover these edges, at least n vertices will be needed, since no two of the edges will share a vertex.

Figure: Reducing satisfiability instance

to vertex cover

Second, we translate the clauses of the 3-SAT problem. For each of the c clauses, we create three new vertices, one for each literal in each clause. The three vertices of each clause will be connected so as to form c triangles. At least two vertices per triangle must be included in any vertex cover of these triangles. Finally, we will connect these two sets of components together. Each literal in the vertex ``gadgets'' is connected to corresponding vertices in the clause gadgets (triangles) that share the same literal. From a 3SAT instance with n variables and c clauses, this constructs a graph with 2n+3c vertices. The complete

file:///E|/BOOK/BOOK3/NODE115.HTM (1 of 2) [19/1/2003 1:29:53]

Vertex Cover

reduction for the 3-SAT problem

is shown in Figure

.

This graph has been designed to have a vertex cover of size n+2c if and only if the original expression is satisfiable. By the earlier analysis, any vertex cover must have at least n+2c vertices, since adding extra edges to the graph can only increase the size of the vertex cover. To show that our reduction is correct, we must demonstrate that: ●

●

Every satisfying truth assignment gives a vertex cover - Given a satisfying truth assignment for the clauses, select the n vertices from the vertex gadgets that correspond to literals to be members of the vertex cover. Since this is a satisfying truth assignment, a true literal from each clause will have covered one of the three cross edges connecting each clause triangle to a vertex gadget. Therefore, by selecting the other two vertices of each clause triangle, we can also pick up the remaining cross edges and complete the cover. Every vertex cover gives a satisfying truth assignment - Given any vertex cover C of size n+2c, exactly n of the vertices must belong to the vertex gadgets. Let these first stage vertices define the truth assignment, while the 2c remaining cover vertices must be distributed at two per clause gadget; otherwise a clause gadget edge must go uncovered. These clause gadget vertices can cover only two of the three connecting cross edges per clause. Therefore, if C gives a vertex cover, at least one cross edge per clause must be covered, meaning that the corresponding truth assignment satisfies.

This proof of the hardness of vertex cover, chained with the clique and independent set arguments of Section easier.

, gives us a library of hard graph problems that we can use to make future hardness proofs

Next: Other NP-Complete Problems Up: Difficult Reductions Previous: Integer Programming Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE115.HTM (2 of 2) [19/1/2003 1:29:53]

Other NP-Complete Problems

Next: The Art of Proving Up: Intractable Problems and Approximations Previous: Vertex Cover

Other NP-Complete Problems Clique, vertex cover, and integer programming are just three of the literally hundreds of problems that have been shown to be NP-complete. It is important to be aware of which kinds of problems tend to be hard, so you recognize them when you see them in applications, and also to provide a suitable class of candidates for future reductions. Some, but by no means all, of the hard problems from the catalog include: ●

Integer partition - Can you partition n integers into two subsets such that the sums of the subsets

●

for details. are equal? See Section Bin packing - How many bins of a given size do you need to hold n items of variable size? See

●

for details. Section Chromatic number - How many colors do you need to color a graph such that no neighboring

●

for details. vertices are of the same color? See Section Bandwidth - Which permutation p of the vertices of a graph minimizes the length of the longest edge when the vertices are ordered on a line, i.e. details.

? See Section

for

A few other catalog problems exist in a limbo state, where it is not known whether the problem has a fast algorithm or is NP-complete. The most prominent of these are graph isomorphism (see Section

) and

). That this limbo list is so short is quite a tribute to the state of the art in primality testing (see Section algorithm design and the power of NP-completeness. For almost every important problem for which we do not know a fast algorithm, we have a good solid reason for why one doesn't exist. The same should hold true for the problems you encounter in your work. One way or another they should be resolved as being either hard or polynomial. Leaving them in a limbo state is a sure sign of a bushleague algorithm designer. It takes experience to be able to sense whether a problem is likely to be hard or not. Perhaps the quickest way to gain this experience is through careful study of the catalog. Note that slightly changing the wording of a problem can make the difference between it being polynomial or NP-complete. Finding the shortest path in a graph is easy, while finding the longest path in a graph is hard. Constructing a tour that visits all the edges once in a graph is easy, while constructing a tour that visits all the vertices once is file:///E|/BOOK/BOOK3/NODE116.HTM (1 of 2) [19/1/2003 1:29:54]

Other NP-Complete Problems

hard. The first thing to do when you suspect a problem might be NP-complete is look in Garey and Johnson's book Computers and Intractability [GJ79], which contains a list of several hundred problems known to be NP-complete. Likely you will find the problem you are interested in.

Next: The Art of Proving Up: Intractable Problems and Approximations Previous: Vertex Cover Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE116.HTM (2 of 2) [19/1/2003 1:29:54]

The Art of Proving Hardness

Next: War Story: Hard Against Up: Intractable Problems and Approximations Previous: Other NPComplete Problems

The Art of Proving Hardness Proving that problems are hard is a skill. But once you get the hang of it, reductions can be surprisingly straightforward and pleasurable to do. Indeed, the dirty little secret of NP-completeness proofs is that they are usually easier to create than explain, in the same way that it is often easier to rewrite old code than it is to understand and modify it. I offer the following advice to those seeking to prove the hardness of a given problem: ●

Make your source problem as simple (i.e. restricted) as possible. Never try to use the general traveling salesman problem (TSP) as a source problem. Better, use Hamiltonian cycle, i.e. TSP where all the weights are 1 or . Even better, use Hamiltonian path instead of cycle, so you don't have to worry about closing up the path. Best of all, use Hamiltonian path on directed planar graphs where each vertex has total degree 3. All of these problems are equally hard, and the more you can restrict the problem that you are reducing, the less work your reduction has to do.

●

As another example, never try to use full satisfiability to prove hardness. Start with 3satisfiability. In fact, you don't even have to use full 3-satisfiability. Instead, consider planar 3satisfiability, where there is a way to draw the clauses as a graph in the plane such you can connect all instances of the same literal together without edges crossing. This property tends to be useful in proving the hardness of geometric problems. All these problems are equally hard, and hence NP-completeness reductions using any of them are equally convincing. Make your target problem as hard as possible.

●

Don't be afraid to add extra constraints or freedoms in order to make your problem more general. Perhaps your undirected graph problem can be generalized into a directed graph problem and can hence be easier to prove hard. Once you have a proof of hardness for the general problem, you can then go back and try to simplify the target. Select the right source problem for the right reason. Selecting the right source problem makes a big difference in how difficult it is to prove a problem hard. Although theoretically any particular problem is as good as another, this is the first and

file:///E|/BOOK/BOOK3/NODE117.HTM (1 of 3) [19/1/2003 1:29:55]

The Art of Proving Hardness

easiest place to go wrong. When faced with trying to prove that a problem is hard, some people fish around through dozens of problems, looking for the one that seems the best fit. These people are amateurs; odds are they never will recognize what they are looking for when they see it. I use four and only four problems as candidates for my hard source problem. Limiting them to four means that I can know a lot about each of these problems, such as which variants of these problems are hard and which are soft. My favorite problems are: ❍

❍

❍

❍

3-SAT: The old reliable. When none of the three problems below seem appropriate, I go back to the original source. Integer partition: This is the one and only choice for problems whose hardness seems to require using large numbers. Vertex cover: This is the answer for any graph problems whose hardness depends upon selection. Chromatic number, clique, and independent set all involve trying to select the correct subset of vertices or edges. Hamiltonian path: This is my choice for any graph problem whose hardness depends upon ordering. If you are trying to route or schedule something, Hamiltonian path is likely your lever into the problem.

●

Amplify the penalties for making the undesired selection.

●

Many people are too timid in their thinking about hardness proofs. You are trying to translate one problem into another, while keeping the problems as close to their original identities as as possible. The easiest way to do this is to be bold with your penalties, to punish anyone for trying to deviate from your intended solution. Your thinking should be, ``if you select this element, then you have to pick up this huge set that prevents you from finding an optimal solution.'' The sharper the consequences for doing what is undesired, the easier it is to prove if and only if. Think strategically at a high level, then build gadgets to enforce tactics.

●

You should be asking yourself the following types of questions. ``How can I force that either A or B but not both are chosen?'' ``How can I force that A is taken before B?'' ``How can I clean up the things I did not select?'' After you have an idea of what you want your gadgets to do, you can worry about how to actually craft them. When you get stuck, alternate between looking for an algorithm or a reduction. Sometimes the reason you cannot prove hardness is that there exists an efficient algorithm to solve your problem! Techniques such as dynamic programming or reducing to polynomial-time graph problems such as matching or network flow sometimes yield surprising polynomial algorithms. Whenever you can't prove hardness, it likely pays to alter your opinion occasionally to keep yourself honest.

file:///E|/BOOK/BOOK3/NODE117.HTM (2 of 3) [19/1/2003 1:29:55]

The Art of Proving Hardness

Next: War Story: Hard Against Up: Intractable Problems and Approximations Previous: Other NPComplete Problems Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE117.HTM (3 of 3) [19/1/2003 1:29:55]

War Story: Hard Against the Clock

Next: Approximation Algorithms Up: Intractable Problems and Approximations Previous: The Art of Proving

War Story: Hard Against the Clock My class's attention span was running down like sand through an hourglass. Eyes were starting to glaze even in the front row. Breathing had become soft and regular in the middle of the room. Heads were tilted back and eyes shut in the rear. There were fifteen minutes left to go in my lecture on NP-completeness, and I couldn't really blame them. They had already seen several reductions like the ones presented here, but NP-completeness reductions are easier to create than to understand or explain. They had to watch one being created in order to appreciate this. I reached for my trusty copy of Garey and Johnson's book [GJ79], which contains a list of over four hundred different known NP-complete problems in an appendix in the back. ``Enough of this!'' I announced loudly enough to startle those in the back row. ``NP-completeness proofs are routine enough that we can construct them on demand. I need a volunteer with a finger. Can anyone help me?'' A few students in the front held up their hands. A few students in the back held up their fingers. I opted for one from the front row. ``Select a problem at random from the back of this book. I can prove the hardness of any of these problems in the now twelve minutes remaining in this class. Stick your finger in and read me a problem.'' I had definitely gotten their attention. But I could have done that by offering to juggle chainsaws. Now I had to deliver results without cutting myself into ribbons. The student picked out a problem. ``OK, prove that Inequivalence of Programs with Assignments is hard,'' she said. ``Huh? I've never heard of that problem before. What is it? Read me the entire description of the problem so I can write it on the board.'' The problem was as follows:

file:///E|/BOOK/BOOK3/NODE118.HTM (1 of 4) [19/1/2003 1:29:56]

War Story: Hard Against the Clock

Input: A finite set X of variables, two programs

where the

and

, each a sequence of assignments of the form

are in X; and a value set V.

Output: Is there an initial assignment of a value from V to each variable in X such that the two programs yield different final values for some variable in X? I looked at my watch. Ten minutes to go. But now everything was on the table. I was faced with a language problem. The input was two programs with variables, and I had to test to see whether they always do the same thing. ``First things first. We need to select a source problem for our reduction. Do we start with integer partition? 3-satisfiability? Vertex cover or Hamiltonian path?'' Since I had an audience, I tried thinking out loud. ``Our target is not a graph problem or a numerical problem, so let's start thinking about the old reliable: 3-satisfiability. There seem to be some similarities. 3-SAT has variables. This thing has variables. To be more like 3-SAT, we could try limiting the variables in this problem so they only take on two values, i.e. . Yes. That seems convenient.'' My watch said nine minutes left. ``So, class, which way does the reduction go. 3-SAT to language or language to 3-SAT?'' The front row correctly murmured, ``3-SAT to language.'' ``Right. So we have to translate our set of clauses into two programs. How can we do that? We can try to split the clauses into two sets and write separate programs for each of them. But how do we split them? I don't see any natural way how to do it, because eliminating any single clause from the problem might suddenly make an unsatisfiable formula satisfiable, thus completely changing the answer. Instead, let's try something else. We can translate all the clauses into one program, and then make the second program be trivial. For example, the second program might ignore the input and always outputs either only true or only false. This sounds better. Much better.'' I was still talking out loud to myself, which wasn't that unusual. But I had people listening to me, which was. ``Now, how can we turn a set of clauses into a program? We want to know whether the set of clauses can be satisfied, or if there is an assignment of the variables such that it is true. Suppose we constructed a program to evaluate whether is satisfied. We can do it like this ....''

file:///E|/BOOK/BOOK3/NODE118.HTM (2 of 4) [19/1/2003 1:29:56]

War Story: Hard Against the Clock

It took me a few minutes worth of scratching before I found the right program. I assumed that I had access to constants for and , which seemed reasonable, in the sense that it shouldn't make the problem algorithmically harder. Once my proof worked, I could later worry about removing the extra assumption if desired.

``Great. Now I have a way to evaluate the truth of each clause. I can do the same thing to evaluate whether all the clauses are satisfied.''

Now the back of the classroom was getting excited. They were starting to see a ray of hope that they would get to leave on time. There were two minutes left in class. ``Great. So now we have a program that can evaluate to be true if and only if there is a way to assign the variables so as to satisfy the set of clauses. We need a second program to finish the job. What about ? Yes, that is all we need. Our language problem asks whether the two programs always output the same thing, regardless of the possible variable assignments. If the clauses are satisfiable, that means that there must be an assignment of the variables such that the long program would output true. Testing whether the programs are equivalent is exactly the same as asking if the clauses are satisfiable.'' I lifted my arms in triumph. ``And so, the problem is neat, sweet, and NP-complete.'' I got the last word out just before the bell rang. This exercise was so much fun that I repeated it the next time I taught the course, using a different randomly selected problem. The audio from my ultimately successful attempt to prove hardness is accessible from the CD-ROM, if you want to hear what the creation of an NP-completeness proof sounds file:///E|/BOOK/BOOK3/NODE118.HTM (3 of 4) [19/1/2003 1:29:56]

War Story: Hard Against the Clock

like.

Next: Approximation Algorithms Up: Intractable Problems and Approximations Previous: The Art of Proving Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE118.HTM (4 of 4) [19/1/2003 1:29:56]

Approximation Algorithms

Next: Approximating Vertex Cover Up: Intractable Problems and Approximations Previous: War Story: Hard Against

Approximation Algorithms For the practical person, demonstrating that a problem is NP-complete is never the end of the line. Presumably, there was a reason why you wanted to solve it in the first place. That reason for wanting the solve it will not have gone away on being told that there is no polynomial-time algorithm. You still seek a program that solves the problem of interest. All you know is that you won't find one that quickly solves the problem to optimality in the worst case. You still have the following options: ●

●

●

Algorithms fast in the average case - Examples of such algorithms include backtracking algorithms with substantial pruning. Heuristics - Heuristic methods like simulated annealing or greedy approaches can be used to find a solution with no requirement that it be the best one. Approximation algorithms - The theory of NP-completeness does not stipulate that it is hard to get close to the answer, only that it is hard to get the optimal answer. With clever, problem-specific heuristics, we can often get provably close to the optimal answer.

Approximation algorithms return solutions with a guarantee attached, namely that the optimal solution can never be much better than this given solution. Thus you can never go too far wrong in using an approximation algorithm. No matter what your input instance is and how lucky you are, you are doomed to do all right. Further, approximation algorithms realizing provably good bounds often are conceptually simple, very fast, and easy to program. One thing that is usually not clear, however, is how well the solution from an approximation algorithm compares to what you might get from a heuristic that gives you no guarantees. The answer could be worse or it could be better. Leaving your money in a savings account in a bank guarantees you 3% interest without risk. Still, you likely will do much better putting your money in stocks than in the bank, even though performance is not guaranteed. One way to get the best of approximation algorithms and heuristics is to run both of them on the problem instance and pick the solution giving the better answer. This way, you get a solution that comes with a guarantee and a second chance to do even better. When it comes to heuristics for hard problems, sometimes you can have it both ways.

file:///E|/BOOK/BOOK3/NODE119.HTM (1 of 2) [19/1/2003 1:29:57]

Approximation Algorithms

● ●

Approximating Vertex Cover The Euclidean Traveling Salesman

Next: Approximating Vertex Cover Up: Intractable Problems and Approximations Previous: War Story: Hard Against Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE119.HTM (2 of 2) [19/1/2003 1:29:57]

Approximating Vertex Cover

Next: The Euclidean Traveling Salesman Up: Approximation Algorithms Previous: Approximation Algorithms

Approximating Vertex Cover

As we have seen before, finding the minimum vertex cover of a graph is NP-complete. However, a very simple procedure can efficiently find a cover that is at most twice as large as the optimal cover:

VertexCover(G=(V,E))

while

do:

Select an arbitrary edge

Add both u and v to the vertex cover

Delete all edges from E that are incident on either u or v.

It should be apparent that this procedure always produces a vertex cover, since each edge is only deleted immediately after an incident vertex has been added to the cover. More interesting is the claim that any vertex cover must use at least half as many vertices as this one. Why? Consider just the edges selected by the algorithm. No two of these edges can share a vertex. Therefore, any cover of just these edges must include at least one vertex per edge, which makes it at least half the size of the greedy cover. There are several interesting things to notice about this algorithm: ●

Although the procedure is simple, it is not stupid - Many seemingly smarter heuristics can give a far worse performance in the worst case. For example, why not modify the procedure above to select only one of the two vertices for the cover instead of both. After all, the selected edge will be equally well covered by only one vertex. . This heuristic will produce a two-vertex cover, while the However, consider the star-shaped graph of Figure single vertex heuristic can return a cover as large as n-1 vertices, should we get unlucky and repeatedly select the leaf instead of the center as the cover vertex.

file:///E|/BOOK/BOOK3/NODE120.HTM (1 of 2) [19/1/2003 1:29:58]

Approximating Vertex Cover

Figure: Neglecting to pick the center vertex leads to a terrible vertex cover ●

●

●

Greedy isn't always the answer - Perhaps the most natural heuristic for this problem would repeatedly select and delete the vertex of highest remaining degree for the vertex cover. After all, this vertex will cover the largest number of possible edges. However, in the case of ties or near ties, this heuristic can go seriously astray and in the worst case times optimal. can yield a cover that is Making a heuristic more complicated does not necessarily make it better - It is easy to complicate heuristics by adding more special cases or details. For example, the procedure above does not specify which edge should be selected next. It might seem reasonable always to select the edge whose endpoints have highest degree. However, this does not improve the worst-case bound and just makes it more difficult to analyze. A postprocessing cleanup step can't hurt - The flip side of designing simple heuristics is that they can often be modified to yield better-in-practice solutions without weakening the approximation bound. For example, a postprocessing step that deletes any unnecessary vertex from the cover can only improve things in practice, even though it won't help the worst-case bound.

The important property of approximation algorithms is relating the size of the solution produced directly to a lower bound on the optimal solution. Instead of thinking about how well we might do, we have to think about the worst case i.e. how badly we might perform.

Next: The Euclidean Traveling Salesman Up: Approximation Algorithms Previous: Approximation Algorithms Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE120.HTM (2 of 2) [19/1/2003 1:29:58]

The Euclidean Traveling Salesman

Next: Exercises Up: Approximation Algorithms Previous: Approximating Vertex Cover

The Euclidean Traveling Salesman

In most natural applications of the traveling salesman problem, direct routes are inherently shorter than indirect routes. For example, if the edge weights of the graph are ``as the crow flies'', straight-line distances between pairs of cities, the shortest path from x to y will always be to fly directly.

Figure: The triangle inequality typically holds in geometric and weighted graph problems. The edge weights induced by Euclidean geometry satisfy the triangle inequality, which insists that for all triples of vertices u, v, and w. The reasonableness of this condition is shown in Figure . Note that the cost of airfares is an example of a distance function that violates the triangle inequality, since it is sometimes cheaper to fly through an intermediate city than to fly to the destination directly. TSP remains hard when the distances are Euclidean distances in the plane. Whenever a graph obeys the triangle inequality, we can approximate the optimal traveling salesman tour using minimum spanning trees. First, observe that the weight of a minimum spanning tree is a lower bound on the cost of the optimal tour. Why? Deleting any edge from a tour leaves a path, the total weight of which must be no greater than that of the original tour. This path has no cycles, and hence is a tree, which means its weight is at least that of the minimum spanning tree. Thus the minimum spanning tree cost gives a lower bound on the optimal tour. file:///E|/BOOK/BOOK3/NODE121.HTM (1 of 3) [19/1/2003 1:29:59]

The Euclidean Traveling Salesman

Consider now what happens in performing a depth-first traversal of a spanning tree. Suppose we walk through each tree edge as we process it in a depth-first search. We will visit each edge twice, once going down the tree when exploring it and once going up after exploring the entire subtree. For example, in the , we visit the vertices in order 1-2-1-3-5-8-5-9-5-3-6-3-1-4-7-10-7-11-7-4depth-first search of Figure 1, thus using every tree edge exactly twice. Therefore, this tour has weight twice that of the minimum spanning tree, and hence at most twice optimal.

Figure: A depth-first traversal of a spanning tree, with the shortcut tour However, vertices will be repeated on this depth-first search tour. To remove the extra vertices, at each step we can take a shortest path to the next unvisited vertex. The shortcut tour for the tree above is 1-2-35-8-9-6-4-7-10-11-1. Because we have replaced a chain of edges by a single direct edge, the triangle inequality ensures that the tour can only get shorter. Thus the shortcut tour is within weight twice that of optimal. More complicated but better approximation algorithms for Euclidean TSP are mentioned in Section

. No approximation algorithms exist for TSPs that do not satisfy the triangle inequality.

file:///E|/BOOK/BOOK3/NODE121.HTM (2 of 3) [19/1/2003 1:29:59]

The Euclidean Traveling Salesman

Next: Exercises Up: Approximation Algorithms Previous: Approximating Vertex Cover Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE121.HTM (3 of 3) [19/1/2003 1:29:59]

Exercises

Next: Implementation Challenges Up: Intractable Problems and Approximations Previous: The Euclidean Traveling Salesman

Exercises 1. Prove that the vertex cover problem (does there exist a subset S of k vertices in a graph G such that every edge in G is incident upon at least one vertex in S?) remains NP-complete even when all the vertices in the graph are restricted to have even degree. 2. An instance of the set cover problem consists of a set X of n elements, a family F of subsets of X, and an integer k. The question is, do there exist k subsets from F whose union is X? For example, if does for k=3 (for example,

and

, there does not exist a solution for k=2 but there ).

Prove that set cover is NP-complete with a reduction from vertex cover. 3. The baseball card collector problem is as follows. Given packets , each of which contains a subset of packets? that year's baseball cards, is it possible to collect all the year's cards by buying For example, if the players are

and the packets are

there does not exist a solution for k=2 but there does for k=3, such as

Prove that the baseball card collector problem is NP-hard using a reduction from vertex cover. 4. (*) An Eulerian cycle is a tour that visits every edge in a graph exactly once. An Eulerian subgraph is a subset of the edges and vertices of a graph that has an Eulerian cycle. Prove that the problem of finding the number of edges in the largest Eulerian subgraph of a graph is NP-hard. (Hint: the Hamiltonian circuit problem is NP-hard even if each vertex in the graph is incident upon exactly three edges.) 5. The low degree spanning tree problem is as follows. Given a graph G and an integer k, does G contain a spanning tree such that all vertices in the tree have degree at most k (obviously, only tree edges count towards the degree)? For example, in the following graph, there is no spanning tree such that all vertices have degree less than three.

file:///E|/BOOK/BOOK3/NODE122.HTM (1 of 3) [19/1/2003 1:30:00]

Exercises

1. Prove that the low degree spanning tree problem is NP-hard with a reduction from Hamiltonian path. 2. Now consider the high degree spanning tree problem, which is as follows. Given a graph G and an integer k, does G contain a spanning tree whose highest degree vertex is at least k? In the previous example, there exists a spanning tree of highest degree 8. Give an efficient algorithm to solve the high degree spanning tree problem, and an analysis of its time complexity. 6. (*) The problem of testing whether a graph G contains a Hamiltonian path is NP-hard, where a Hamiltonian path P is a path that visits each vertex exactly once. There does not have to be an edge in G from the ending vertex to the starting vertex of P, unlike in the Hamiltonian cycle problem. Given a directed acyclic graph G (a DAG), give an O(n+m)-time algorithm to test whether or not it contains a Hamiltonian path. (Hint: think about topological sorting and DFS.) 7. (**) The 2-SAT problem is, given a Boolean formula in 2-conjunctive normal form (CNF), decide whether the formula is satisfiable. 2-SAT is like 3-SAT, except that each clause can have only two literals. For example, the following formula is in 2-CNF:

Give a polynomial-time algorithm to solve 2-SAT. 8. (*) It is an open question whether the decision problem ``Is integer n a composite number, in other words, not prime?" can be computed in time polynomial in the size of the input. Why doesn't the following algorithm suffice to prove it is in P, since it runs in O(n) time?

PrimalityTesting(n)

composite :=

for i := 2 to n-1 do

if

then

composite :=

file:///E|/BOOK/BOOK3/NODE122.HTM (2 of 3) [19/1/2003 1:30:00]

Exercises

Next: Implementation Challenges Up: Intractable Problems and Approximations Previous: The Euclidean Traveling Salesman Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE122.HTM (3 of 3) [19/1/2003 1:30:00]

Data Structures

Next: Dictionaries Up: A Catalog of Algorithmic Previous: Caveats

Data Structures Data structures are not really algorithms that you can find and plug into your application. Instead, they are the fundamental constructs for you to build your program around. Becoming fluent in what data structures can do for you is essential to get full value from them. Because of this, this section is slightly out of sync with the rest of the catalog. Perhaps the most useful aspect of it will be the pointers to implementations of various data structures. Many of these data structures are nontrivial to implement well, so the programs we point to will likely be useful as models even if they do not do exactly what you need. Certain fundamental data structures, like kd-trees and suffix trees, are not as well known as they should be. Hopefully, this catalog will serve to better publicize them. There is a large number of books on elementary data structures available. Our favorites include: ●

●

●

Gonnet and Baeza-Yates [GBY91] - The book is a comprehensive reference to fundamental searching, sorting, and text searching algorithms. It features over 2,000 references and implementations in C and Pascal. These programs are now available by ftp/WWW. See Section for more details. Weiss [Wei92] - A nice text, emphasizing data structures more than algorithms. Comes in Pascal, C++, and Ada editions. Wood [Woo93] - A thorough and accessible treatment of modern data structures, including suffix trees and geometric data structures. Pascal implementations for many of the data structures are provided.

Mehlhorn and Tsakalidis [MT90b] provide a detailed and up-to-date survey of research in data structures. The student who took only an elementary course in data structures is likely to be impressed by the volume and quality of recent work on the subject.

●

Dictionaries

file:///E|/BOOK/BOOK3/NODE128.HTM (1 of 2) [19/1/2003 1:30:01]

Data Structures ● ● ● ● ●

Priority Queues Suffix Trees and Arrays Graph Data Structures Set Data Structures Kd-Trees

Next: Dictionaries Up: A Catalog of Algorithmic Previous: Caveats Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE128.HTM (2 of 2) [19/1/2003 1:30:01]

Dictionaries

Next: Priority Queues Up: Data Structures Previous: Data Structures

Dictionaries

Input description: A set of n records, each identified by one or more key fields. Problem description: Build and maintain a data structure to efficiently locate, insert, or delete the record associated with any query key q. Discussion: The abstract data type ``dictionary'' is one of the most important structures in computer science. Dozens of different data structures have been proposed for implementing dictionaries including hash tables, skip lists, and balanced/unbalanced binary search trees - so choosing the right one can be tricky. Depending on the application, it is also a decision that can significantly impact performance. In practice, it is more important to avoid using a bad data structure than to identify the single best option available. An essential piece of advice is to carefully isolate the implementation of the dictionary data structure from its interface. Use explicit calls to subroutines that initialize, search, and modify the data structure, rather than embedding them within the code. This leads to a much cleaner program, but it also makes it easy to try different dictionary implementations to see how they impact performance. Do not obsess about the cost of the procedure call overhead inherent in such an abstraction. If your application is so time-critical that such overhead can impact performance, then it is even more essential that you be able to file:///E|/BOOK/BOOK3/NODE129.HTM (1 of 5) [19/1/2003 1:30:03]

Dictionaries

easily experiment with different implementations of your dictionary. In choosing the right data structure for your dictionary, ask yourself the following questions: ●

●

●

●

How many items will you typically have in your data structure? - Will you know this number in advance? Are you looking at a problem small enough that the simple data structure will be best, or will it be so large that we must worry about using too much memory or swapping? Do you know the relative number of insertions, deletions, and search queries? - Will there be any modifications to the data structure after it is first constructed, or will it be static from that point on? Do you have an understanding of the relative frequency with which different keys will be accessed? - Can we assume that the access pattern will be uniform and random, or will it exhibit a skewed access distribution (i.e. certain elements are much more popular than others) or a sense of locality (i.e. elements are likely to be repeatedly accessed in clusters, instead of at fairly random intervals). Usually, the world is both skewed and clustered. Is it critical that individual operations be fast, or only that the total amount of work done over the entire program be minimized? - When response time is critical, such as in a program controlling a heart-lung machine, you can't wait too long between steps. When you have a program that is doing a lot of queries over the database, such as identifying all sex offenders who happen to be Republicans, it is not so critical that you pick out any particular congressman quickly as that you get them all with the minimum total effort.

Once you understand what your needs are, try to identify the best data structure from the list below: ●

●

●

Unsorted linked lists or arrays - For small data sets, say up to 10 to 20 items, an unsorted array is probably the easiest and most efficient data structure to maintain. They are easier to work with than linked lists, and if the dictionary will be kept this small, you cannot possibly save a significant amount of space over allocating a full array. If your dictionary will be too much larger, the search time will kill you in either case. A particularly interesting and useful variant is a self-organizing list. Whenever a key is accessed or inserted, always move it to head of the list. Thus if the key is accessed again in the near future, it will be near the front and so require only a short search to find it. Since most applications exhibit both uneven access frequencies and locality of reference, the average search time for a successful search in a self-organizing list is typically much better than in a sorted or unsorted list. Of course, self-organizing data structures can be built from arrays as well as linked lists. Sorted linked lists or arrays - Maintaining a sorted linked list is usually not worth the effort (unless you are trying to eliminate duplicates), since we cannot perform binary search in such a data structure. A sorted array will be appropriate if and only if there are not many insertions or deletions. When the array gets so large that it doesn't fit in real memory, think B-trees instead. Hash tables - For applications involving a moderate-to-large number of keys (say between 100 and 1,000,000), a hash table with bucketing is probably the right way to go. In a hash table, we

file:///E|/BOOK/BOOK3/NODE129.HTM (2 of 5) [19/1/2003 1:30:03]

Dictionaries

use a function that maps keys (be they strings, numbers, or whatever) to integers between 0 and m1. We maintain an array of m buckets, each typically implemented using an unsorted linked list. For a given key, the hash function immediately identifies which bucket will contain it. If we use a hash function that spreads the keys out nicely and a sufficiently large hash table, each bucket should contain very few items, thus making linear search acceptable. Insertion and deletion from a hash table reduce to insertion and deletion from the bucket/list. A well-tuned hash table will likely outperform a sorted array in most applications. However, several design decisions go into creating a well-tuned hash table: ❍

❍

How big should the table be? Typically, m should about the same as the maximum number of items you expect to put in the table. Make sure that m is a prime number, so as to minimize the dangers of a bad hash function. What hash function should I use? For strings, something like

should work, where is the size of the alphabet and char(x) is the function that maps each character x to its ASCII character code. For long strings, 8 to 10 characters should be sufficient to hash upon, provided they are unlikely to be padded blanks or some other invariant. Use Horner's rule to implement this hash function computation efficiently, as discussed in Section

●

.

Regardless of which hash function you decide to use, print statistics on the distribution of keys per bucket to see how uniform it really is. Odds are the first hash function you try will not prove to be the best. Botching up the hash function is an excellent way to slow down any application. Binary search trees - Binary search trees are elegant data structures that support fast insertions, deletions, and queries. The big distinction between different types of trees is whether they are explicitly rebalanced after insertion or deletion, and how this rebalancing takes place. In random search trees, no rebalancing takes place and we simply insert a node at the leaf position where we can find it. Although search trees perform well under random insertions, most applications are not really random. Indeed, unbalanced search trees constructed by inserting keys in sorted order are a disaster, performing like a linked list. Balanced search trees use local rotation operations to restructure search trees, moving more distant nodes closer to the root while maintaining the in-order search structure of the tree. Among balanced search trees, AVL and 2/3 trees are now passé, and red-black trees seem to be more popular. A particularly interesting self-organizing data structure is the splay tree, which uses rotations to move any accessed key to the root. Frequently used or recently accessed nodes thus sit near the top of the tree, allowing fast search.

file:///E|/BOOK/BOOK3/NODE129.HTM (3 of 5) [19/1/2003 1:30:03]

Dictionaries

●

Bottom line: Which binary search tree is best for your application? Probably the balanced tree for which you have the best implementation readily available. See the choices below. Which flavor of balanced tree is probably not as important as how good the programmer was who coded it. B-trees - For data sets so large that they will not fit in main memory (say more than 1,000,000 items) your best bet will be some flavor of a B-tree. As soon as the data structure gets outside of main memory, the search time to access a particular location on a disk or CD-ROM can kill you, since this is several orders of magnitude slower than accessing RAM. The idea behind a B-tree is to collapse several levels of a binary search tree into a single large node, so that we can make the equivalent of several search steps before another disk access is needed. We can thereafter reference enormous numbers of keys using only a few disk accesses. To get the full benefit from using a B-tree, it is important to understand explicitly how the secondary storage device and virtual memory interact, through constants such as page size and virtual/real address space.

●

Even for modest-sized data sets, unexpectedly poor performance of a data structure may be due to excessive swapping, so listen to your disk to help decide whether you should be using a B-tree. Skip lists - These are somewhat of a cult data structure. Their primary benefits seem to be ease of implementation relative to balanced trees. If you are using a canned tree implementation, and thus not coding it yourself, this benefit is eliminated. I wouldn't bother with them.

Implementations: LEDA (see Section ) provides an extremely complete collection of dictionary data structures in C++, including hashing, perfect hashing, B-trees, red-black trees, random search trees, and skip lists. Given all of these choices, their default dictionary implementation is a randomized search tree [AS89], presumably reflecting which structure they expect to be most efficient in practice. ) is an algorithm animation system for UNIX and X-windows that includes XTango (see Section animations of such dictionary data structures as AVL trees, binary search trees, hashing, red-black trees, and treaps (randomized search trees). Many of these are interesting and quite informative to watch. Further, the C source code for each animation is included. The 1996 DIMACS implementation challenge focused on elementary data structures like dictionaries. The world's best available implementations were likely to be identified during the course of the challenge, and they are accessible from http://dimacs.rutgers.edu/ . Bare bones implementations in C and Pascal of a dizzying variety of dictionary data structures appear in [GBY91], among them several variations on hashing and binary search trees, and optimal binary search tree construction. See Section

for details.

file:///E|/BOOK/BOOK3/NODE129.HTM (4 of 5) [19/1/2003 1:30:03]

Dictionaries

Implementation-oriented treatments of a variety of dictionary data structures appear in [BR95], including hashing, splay trees, red-black trees, and what looks like a thorough implementation of B-trees. Code in C for these data structures is included in the text and is available on disk for a modest fee. Notes: Mehlhorn and Tsakalidis [MT90b] give a thorough survey of the state of the art in modern data structures. Knuth [Knu73a] provides a detailed analysis and exposition on fundamental dictionary data structures but misses such modern data structures as red-black and splay trees. Gonnet and Baeza-Yates [GBY91] provide implementations (in C and Pascal), detailed references, and experimental results for a wide variety of dictionary data structures. We defer to these sources to avoid giving original references for each of the data structures described above. Good expositions on red-black trees [GS78] include [BR95, CLR90, Woo93]. Good expositions on splay trees [ST85] include [Tar83, Woo93]. Good expositions on B-trees [BM72] include [BR95, CLR90]. Good expositions on hashing includes [Meh84, Woo93]. Several modern data structures, such as splay trees, have been studied via amortized analysis, where we bound the total amount of time used by any sequence of operations. In an amortized analysis, we show that if a single operation is very expensive, this is because we have already benefited from enough cheap operations before it to pay off the higher cost. A data structure realizing an amortized complexity of O(f(n)) is less desirable than one whose worst-case complexity is O(f(n)) (since a very bad operation might still occur) but better than one with an average-case complexity O(f(n)), since the amortized bound will achieve this average on any input. Newer dictionary data structures that explicitly incorporate randomization into the construction include randomized search trees [AS89] and skip lists [Pug90]. Related Problems: Sorting (see page

), Searching (see page

).

Next: Priority Queues Up: Data Structures Previous: Data Structures Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE129.HTM (5 of 5) [19/1/2003 1:30:03]

Priority Queues

Next: Suffix Trees and Arrays Up: Data Structures Previous: Dictionaries

Priority Queues

Input description: A set of records with numerically or otherwise totally ordered keys. Problem description: Build and maintain a data structure for quickly inserting and deleting records, while enabling quick access to the smallest or largest key in the set. Discussion: Priority queues are useful data structures in simulations, particularly for maintaining a set of future events ordered by time so that we can quickly retrieve what the next thing to happen is. They are called ``priority'' queues because they enable you to retrieve items not by the insertion time (as in a stack or queue), nor by a key match (as in a dictionary), but by which item has the highest priority of retrieval.

If your application performs no insertions after the first query, there is no need for an explicit priority queue. Simply sort the records by priority and proceed from top to bottom, maintaining a pointer to the last record deleted. This situation occurs in Kruskal's minimum spanning tree algorithm, or when simulating a completely scripted set of events. However, if you are mixing insertions, deletions, and queries, you need a real priority queue. The following questions will help select the right one: file:///E|/BOOK/BOOK3/NODE130.HTM (1 of 4) [19/1/2003 1:30:05]

Priority Queues

●

●

●

Besides access to the smallest element, what other operations will you need? - Will you be searching for arbitrary keys, or just searching for the smallest? Will you be deleting arbitrary elements from the data, or just repeatedly deleting the top or smallest element? Will you know the maximum size of your data structure in advance, or might an arbitrary number of items be inserted into it? - The issue here is whether you can preallocate space for the data structure. Will you be changing the priority of elements already in the queue, or simply inserting and removing them? - Changing the priority of elements implies that we must be able to look up elements in the queue based on their key, in addition to being able to retrieve the largest element.

Depending upon the answers, you have the following basic priority queue choices: ●

●

●

●

●

Sorted array or list - In a sorted array, it is very efficient to find and (by decrementing the top index) delete the smallest element. However, maintaining sortedness makes the insertion of new elements slow. Sorted arrays are suitable when there will be few insertions into the priority queue. Binary heaps - This simple, elegant data structure supports both insertion and extract-min in time each. Heaps maintain an implicit binary tree structure in an array, such that the key of the root of any subtree is less than that of all its descendents. Thus the minimum key is always at the root of the heap. New keys can be inserted by placing them at an open leaf and percolating the element upwards until it sits at its proper place in the partial order. Binary heaps are the right answer whenever you know an upper bound on the number of items in your priority queue, since you must specify the array size at creation time. Bounded height priority queue - This array-based data structure permits constant-time insertion and find-min operations whenever the range of possible key values is limited. Suppose we know that all key values will be integers between 1 and n. We can set up an array of n linked lists, such that the ith list serves as a bucket containing all items with key i. We will maintain a pointer top to the smallest nonempty list. To insert an item with key k into the priority queue, add it to the kth bucket and set . To extract the minimum, report the first item from bucket top, delete it, and move top down if the bucket is now empty. Bounded height priority queues are very useful in maintaining the vertices of a graph sorted by degree, which is a fundamental operation in graph algorithms. Still, they are not as widely known as they should be. They are usually the right priority queue for any small, discrete range of keys. Binary search trees - Binary search trees make effective priority queues, since the smallest element is always the leftmost leaf, while the largest element is always the rightmost leaf. The min (max) is found by simply tracing down left (right) pointers until the next pointer is nil. Binary tree heaps prove most appropriate when you also need to search a dictionary of the values, or if you have an unbounded key range and do not know the maximum priority queue size in advance. Fibonacci and pairing heaps - These complicated priority queues are designed to speed up decrease-key operations, where the priority of an item already in the priority queue is reduced.

file:///E|/BOOK/BOOK3/NODE130.HTM (2 of 4) [19/1/2003 1:30:05]

Priority Queues

This arises, for example, in shortest path computations whenever we discover a shorter route to a vertex v than we had previously established. Thus v has a higher priority of being accessed next.

Properly implemented and used, they lead to better performance on very large computations. Still, they are sufficiently complicated that you shouldn't mess with them unless you really know what you are doing. Implementations: LEDA (see Section ) provides a complete collection of priority queues in C++, including Fibonacci heaps, pairing heaps, Emde-Boas trees, and bounded height priority queues. Fibonacci heaps are their default implementation. SimPack/Sim++ is a library of routines for implementing discrete event simulations, built by Robert Cubert and Paul Fishwick, of the University of Florida. Priority queues are integral to such simulations, and Sim++ contains implementations of linked, binary, leftist, and calendar heaps [Bro88]. If you need a priority queue to control a simulation, check out http://www.cis.ufl.edu/ fishwick/simpack/simpack.html. An associated book [Fis95] describes model design using SimPack. Bare bones implementations in C and Pascal of the basic priority queue data structures appear in [GBY91]. Most notable is the inclusion of implementations of exotic priority queues such as P-trees and pagodas. See Section

for further details.

) is an algorithm animation system for UNIX and X-windows, that includes XTango (see Section animations of such advanced priority queue data structures as binomial and Fibonacci heaps, as well as a spiffy animation of heapsort. Many textbooks provide implementations of simple priority queues, including [MS91] (see Section ). Algorithm 561 [Kah80] of the Collected Algorithms of the ACM is a Fortran implementation of a heap (see Section

).

Notes: Good expositions on efficient heap construction algorithms include [Baa88, Ben86, CLR90, Man89, MT90b]. See [GBY91] for a description of several exotic priority queues. Empirical comparisons between priority queue data structures include [Jon86]. Bounded height priority queues are useful data structures in practice, but they do not have good worstcase performance bounds when arbitrary insertions and deletions are permitted. However, von Emde Boas priority queues [vEBKZ77] support insertion, deletion, search, max, and min operations where each key is an element from 1 to n.

file:///E|/BOOK/BOOK3/NODE130.HTM (3 of 4) [19/1/2003 1:30:05]

Priority Queues

Fibonacci heaps [FT87] support insert and decrease-key operations in O(1) amortized time, with amortized time extract-min and delete operations. The constant-time decrease-key operation leads to faster implementations of classical algorithms for shortest-paths, weighted bipartite-matching, and minimum-spanning-tree. In practice, Fibonacci heaps are nontrivial to implement and have large constant factors associated with them. However, pairing heaps have been proposed to realize the same bounds with less overhead. Experiments with pairing heaps are reported in [SV87]. Heaps define a partial order that can be built using a linear number of comparisons. The familiar lineartime merging algorithm for heap construction is due to Floyd [Flo64]. In the worst case, 1.625n comparisons are necessary [CC92]. comparisons suffice [GM86] and Related Problems: Dictionaries (see page

), sorting (see page

), shortest path (see page

Next: Suffix Trees and Arrays Up: Data Structures Previous: Dictionaries Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE130.HTM (4 of 4) [19/1/2003 1:30:05]

).

Suffix Trees and Arrays

Next: Graph Data Structures Up: Data Structures Previous: Priority Queues

Suffix Trees and Arrays

Input description: A reference string S. Problem description: Build a data structure for quickly finding all places where an arbitrary query string q is a substring of S. Discussion: Suffix trees and arrays are phenomenally useful data structures for solving string problems efficiently and with elegance. If you need to speed up a string processing algorithm from to linear time, proper use of suffix trees is quite likely the answer. Indeed, suffix trees are the hero of the war story . In its simplest instantiation, a suffix tree is simply a trie of the n strings that reported in Section are suffixes of an n-character string S. A trie is a tree structure, where each node represents one character, and the root represents the null string. Thus each path from the root represents a string, described by the characters labeling the nodes traversed. Any finite set of words defines a trie, and two words with common prefixes will branch off from each other at the first distinguishing character. Each leaf represents the end of a string. Figure

illustrates a simple trie.

file:///E|/BOOK/BOOK3/NODE131.HTM (1 of 4) [19/1/2003 1:30:07]

Suffix Trees and Arrays

Figure: A trie on strings the, their, there, was, and when Tries are useful for testing whether a given query string q is in the set. Starting with the first character of q, we traverse the trie along the branch defined by the next character of q. If this branch does not exist in the trie, then q cannot be one of the set of strings. Otherwise we find q in |q| character comparisons regardless of how many strings are in the trie. Tries are very simple to build (repeatedly insert new strings) and very fast to search, although they can be expensive in terms of memory. A suffix tree is simply a trie of all the proper suffixes of S. The suffix tree enables you to quickly test whether q is a substring of S, because any substring of S is the prefix of some suffix (got it?). The search time is again linear in the length of q. The catch is that constructing a full suffix tree in this manner can require time and, even worse, space, since the average length of the n suffices is n/2 and there is likely to be relatively little overlap representing shared prefixes. However, linear space suffices to represent a full suffix tree by being clever. Observe that most of the nodes in a trie-based suffix tree occur on simple paths between branch nodes in the tree. Each of these simple paths corresponds to a substring of the original string. By storing the original string in an array and collapsing each such path into a single node described by the starting and ending array indices representing the substring, we have all the information of the full suffix tree in only O(n) space. The output figure for this section displays a collapsed suffix tree in all its glory. Even better, there exist linear-time algorithms to construct this collapsed tree that make clever use of pointers to minimize construction time. The additional pointers used to facilitate construction can also be used to speed up many applications of suffix trees. But what can you do with suffix trees? Consider the following applications. For more details see the books by Gusfield [Gus97] or Crochemore and Rytter [CR94]: file:///E|/BOOK/BOOK3/NODE131.HTM (2 of 4) [19/1/2003 1:30:07]

Suffix Trees and Arrays

●

●

●

Find all occurrences of q as a substring of S - Just as with a trie, we can walk down from the root to find the node associated with q. The positions of all occurrences of q in S are represented by the descendents of , which can be identified using a depth-first search from . For collapsed suffix trees, this takes O(|q|+k) time if there are k occurrences of q in S. Longest substring common to a set T of strings - Build a single collapsed suffix tree containing all suffixes of all strings, with each leaf labeled with its original string. In the course of doing a depth-first search on this tree, we can label each node with both the length of its common prefix and the number of distinct strings from T that are children of it. Thus the best node can be selected in linear time. Find the longest palindrome in S - A palindrome is a string that reads the same if the order of characters is reversed, such as madam. To find the longest palindrome in a string S, build a single suffix tree containing all suffixes of S and the reversal of S, with each leaf identified by its starting position. A palindrome is defined by any node in this tree that has forward and reversed children from the same position.

Since the linear time suffix tree construction algorithm is tricky, I recommend either starting from an existing implementation or using a simple, potentially quadratic-time incremental-insertion algorithm to build a compressed suffix tree. Another good option is to use suffix arrays, discussed below. Suffix arrays do most of what suffix trees do, while typically using four times less memory than suffix trees. They are also easier to implement. A suffix array is basically just an array that contains all the n suffixes of S in sorted order. Thus a binary search of this array for string q suffices to locate the prefix of a suffix that matches q, permitting efficient substring search in string comparisons. In fact, only character comparisons need be performed on any query, since we can identify the next character that must be tested in the binary search. For example, if the lower range of the search is cowabunga and the upper range is cowslip, all keys in between must share the same first three letters, so only the fourth character of any intermediate key must be tested against q. The space savings of suffix arrays result because as with compressed suffix trees, it suffices to store pointers into the original string instead of explicitly copying the strings. Suffix arrays use less memory than suffix trees by eliminating the need for explicit pointers between suffixes since these are implicit in the binary search. In practice, suffix arrays are typically as fast or faster to search than suffix trees. Some care must be taken to construct suffix arrays efficiently, however, since there are characters in the strings being sorted. A common solution is to first build a suffix tree, then perform an in-order traversal of it to read the strings off in sorted order! Implementations: Ting Chen's and Dimitris Margaritis's C language implementations of suffix trees, reported in the war story of Section , are available on the algorithm repository WWW site: http://www.cs.sunysb.edu/ algorith. Bare bones implementations in C of digital and Patricia trie data structures and suffix arrays appear in file:///E|/BOOK/BOOK3/NODE131.HTM (3 of 4) [19/1/2003 1:30:07]

Suffix Trees and Arrays

[GBY91]. See Section

for details.

Notes: Tries were first proposed by Fredkin [Fre62], the name coming from the central letters of the word ``retrieval''. A survey of basic trie data structures with extensive references appears in [GBY91]. Expositions on tries include [AHU83]. Efficient algorithms for suffix tree construction are due to Weiner [Wei73], McCreight [McC76], and Ukkonen [Ukk92]. Good expositions on these algorithms include Crochmore and Wytter [CR94] and Gusfield [Gus97]. Textbooks include [Woo93] and [AHU74], where they are called position trees. Several applications of suffix trees to efficient string algorithms are discussed in [Apo85]. Suffix arrays were invented by Manber and Myers [MM90], although an equivalent idea called Pat trees due to Gonnet and Baeza-Yates appears in [GBY91]. The power of suffix trees can be further augmented by using a data structure for computing the least common ancestor of any pair of nodes x, y in a tree in constant time, after linear-time preprocessing of the tree. The original data structure is due to Harel and Tarjan [HT84], but it was significantly simplified by Schieber and Vishkin [SV88]. Expositions include Gusfield [Gus97]. The least common ancestor (LCA) of two nodes in a suffix tree or trie defines the node representing the longest common prefix of the two associated strings. Being able to answer such queries in constant time is amazing, and useful as a building block for many other algorithms. The correctness of the LCA data structure is difficult to see; however, it is implementable and can perform well in practice. Related Problems: string matching (see page

), longest common substring (see page

Next: Graph Data Structures Up: Data Structures Previous: Priority Queues Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE131.HTM (4 of 4) [19/1/2003 1:30:07]

).

Graph Data Structures

Next: Set Data Structures Up: Data Structures Previous: Suffix Trees and Arrays

Graph Data Structures

Input description: A graph G. Problem description: Give a flexible, efficient data structure to represent G. Discussion: While there are several possible variations, the two basic data structures for graphs are adjacency matrices and adjacency lists. What these data structures actually are is discussed in Section .

The issues in deciding which one to use include: ●

●

●

How big will your graph be? - How many vertices will it have, both typically and in the worse case? Ditto for the number of edges? If your graph has 100 vertices, your adjacency matrix contains 10,000 entries. If your graph has 1,000 vertices, your adjacency matrix contains 1,000,000 entries. If your graph has 10,000 vertices, your adjacency matrix contains 100,000,000 entries - so forget about it. Adjacency matrices work only for small or very dense graphs. How dense will your graph be? - If the graph is very dense, meaning that a large fraction of the vertex pairs define edges, there is probably no compelling reason to use adjacency lists, since you space, anyway. will be doomed to using Which algorithms will you be implementing? - Certain algorithms are easier on adjacency matrices (such as all-pairs shortest path) and others on adjacency lists (such as most DFS-based

file:///E|/BOOK/BOOK3/NODE132.HTM (1 of 5) [19/1/2003 1:30:08]

Graph Data Structures

●

algorithms). Adjacency matrices win for algorithms that repeatedly ask, ``Is (i,j) in G?'' However, most graph algorithms can be modified to eliminate such queries. Will you be modifying the graph over the course of your application, and if so, how? - Repeated edge insertions and (particularly) deletions argue for adjacency matrices, or perhaps for fancier versions of adjacency lists such as binary search trees. However, more likely than modifying the topology of graph is modifying the attributes of a vertex or edge of the graph, such as size, weight, or color. Attributes are best handled as extra fields in the vertex or edge records of adjacency lists.

Building a good general-purpose graph type is surprisingly tricky and difficult. For this reason, we suggest that you check out existing implementations (particularly LEDA) before hacking up your own. Note that it costs only time linear in the size of the larger data structure to convert between adjacency matrices and adjacency lists. This conversion is unlikely to be the bottleneck in any application, if you decide you want to use both data structures and have the space to store them. This usually isn't necessary but might prove simplest if you are confused about the alternatives. Planar graphs are those that can be drawn in the plane so that no two edges cross. Many graphs arising in applications are planar by definition, such as maps of countries, while others are planar by happenstance, like any tree. Planar graphs are always sparse, since any n-vertex planar graph can have at most 3n-6 edges, so they should usually be represented by adjacency lists. If the planar drawing (or embedding) of the graph is fundamental to what is being computed, planar graphs are best represented geometrically. See Section

for algorithms for constructing planar embeddings from graphs, and

for algorithms maintaining graphs implicit in the arrangements of geometric objects like Section lines and polygons. Hypergraphs are generalized graphs where each edge may link subsets of more than two vertices. For example, suppose we want to represent who is on which Congressional committee. The vertices of our hypergraph would be the individual congressmen, while each hyperedge would represent one committee. Such arbitrary collections of subsets of a set are naturally thought of as hypergraphs. Two basic data structures for hypergraphs are: ●

●

Incidence matrices, which are analogous to adjacency matrices and require space, where m is the number of hyperedges. Each row corresponds to a vertex, and each column to an edge, with a nonzero entry in M[i,j] iff vertex i is incident to edge j. For standard graphs, there are two nonzero entries in each column. The degree of each vertex governs the number of nonzero entries in each row. Bipartite incidence structures, which are analogous to adjacency lists, and hence suited for sparse hypergraphs. There is a vertex of the incidence structure associated with each edge and vertex of the hypergraphs, and an edge (i,j) in the incidence structure if vertex i of the hypergraph is in edge j of the hypergraph. Adjacency lists are typically used to represent this incidence structure. This bipartite incidence structure also provides a natural way to visualize the hypergraph, by drawing

file:///E|/BOOK/BOOK3/NODE132.HTM (2 of 5) [19/1/2003 1:30:08]

Graph Data Structures

the associated bipartite graph. Special efforts must be taken to represent very large graphs efficiently. However, interesting problems have been solved on graphs with millions of edges and vertices. The first step is to make your data ) or structure as lean as possible, by packing your adjacency matrix as a bit vector (see Section removing extra pointers from your adjacency list representation. For example, in a static graph (no edge insertions or deletions) each edge list can be replaced by a packed array of vertex identifiers, thus eliminating pointers and saving potentially half the space. At some point it may become necessary to switch to a hierarchical representation of the graph, where the vertices are clustered into subgraphs that are compressed into single vertices. Two approaches exist for making such a hierarchical decomposition. The first breaks things into components in a natural or application-specific way. For example, knowing that your graph is a map of roads and cities suggests a natural decomposition - partition the map into districts, towns, counties, and states. The other approach runs a graph partition algorithm as in Section . If you are performing the decomposition for space or paging reasons, a natural decomposition will likely do a better job than some naive heuristic for an NPcomplete problem. Further, if your graph is really unmanageably large, you cannot afford to do a very good job of algorithmically partitioning it. You should first verify that standard data structures fail on your problem before attempting such heroic measures. Implementations: LEDA (see Section ) provides the best implemented graph data type currently available in C++. If at all possible, you should use it. If not, you should at least study the methods it provides for graph manipulation, so as to see how the right level of abstract graph type makes implementing algorithms very clean and easy. Although a general graph implementation like LEDA may be 2 to 5 times slower and considerably less space efficient than a stripped-down special-purpose implementation, you have to be a pretty good programmer to realize this performance improvement. Further, this speed is likely to come at the expense of simplicity and clarity. GraphEd [Him94], written in C by Michael Himsolt, is a powerful graph editor that provides an interface for application modules and a wide variety of graph algorithms. If your application demands interaction and visualization more than sophisticated algorithmics, GraphEd might be the right place to start, although it can be buggy. GraphEd can be obtained by anonymous ftp from forwiss.uni-passau.de (132.231.20.10) in directory /pub/local/graphed. See Section graph drawing systems.

for more details on GraphEd and other

) provides a simple but flexible graph data structure in CWEB, a The Stanford Graphbase (see Section literate version of the C language. It is instructive to see what Knuth does and does not place in his basic data structure, although we recommend LEDA as a better basis for further development. LINK is an environment for combinatorial computing that provides special support for hypergraphs, file:///E|/BOOK/BOOK3/NODE132.HTM (3 of 5) [19/1/2003 1:30:08]

Graph Data Structures

including the visualization of hypergraphs. Although written in C++, it provides a Scheme language interface for interacting with the graphs. LINK is available from http://dimacs.rutgers.edu/Projects/LINK.html . An elementary implementation of a ``lazy'' adjacency matrix in Pascal, which does not have to be initialized, appears in [MS91].

See Section

.

Simple graph data structures in Mathematica are provided by Combinatorica [Ski90], with a library of algorithms and display routines.

See Section

.

Notes: It was not until the linear-time algorithms of Hopcroft and Tarjan [HT73b, Tar72] that the advantages of adjacency list data structures for graphs became apparent. The basic adjacency list and matrix data structures are presented in essentially all books on algorithms or data structures, including [CLR90, AHU83, Tar83]. An interesting question concerns minimizing the number of bits needed to represent arbitrary graphs on n vertices, particularly if certain operations must be supported efficiently. Such issues are discussed in [vL90b]. Dynamic graph algorithms are essentially data structures that maintain quick access to an invariant (such as minimum spanning tree or connectivity) under edge insertion and deletion. Sparsification [EGIN92] is a general and interesting approach to constructing dynamic graph algorithms. See [ACI92] for an experimental study on the practicality of dynamic graph algorithms. Hierarchically-defined graphs often arise in VLSI design problems, because designers make extensive use of cell libraries [Len90]. Algorithms specifically for hierarchically-defined graphs include planarity testing [Len89], connectivity [LW88], and minimum spanning trees [Len87]. The theory of hypergraphs is presented by Berge [Ber89]. Related Problems: Set data structures (see page

), graph partition (see page

Next: Set Data Structures Up: Data Structures Previous: Suffix Trees and Arrays Algorithms file:///E|/BOOK/BOOK3/NODE132.HTM (4 of 5) [19/1/2003 1:30:08]

).

Graph Data Structures

Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE132.HTM (5 of 5) [19/1/2003 1:30:08]

Set Data Structures

Next: Kd-Trees Up: Data Structures Previous: Graph Data Structures

Set Data Structures

Input description: A universe of items . where

and a collection of subsets

Problem description: Represent each subset so as to efficiently (1) test whether union or intersection of and , and (3) insert or delete members of S.

,

, (2) find the

Discussion: In mathematical terms, a set is an unordered collection of objects drawn from a fixed universal set. However, it is usually useful for implementation to represent each set in a single canonical order, typically sorted, so as to speed up or simplify various operations. Sorted order turns the problem of finding the union or intersection of two subsets into a linear-time operation - just sweep from left to right and see what you are missing. It also makes possible element searching in sublinear time. Finally, printing the elements of a set in a canonical order paradoxically reminds us that order really doesn't matter. We distinguish sets from two other kinds of objects: strings and dictionaries. If there is no fixed-size universal set, a collection of objects is best thought of as a dictionary, as discussed in Section . If the is not the same as , then your structure is more order does matter in a subset, i.e. if

file:///E|/BOOK/BOOK3/NODE133.HTM (1 of 4) [19/1/2003 1:30:10]

Set Data Structures

profitably thought of as a string, so see Sections

and

.

When each subset has cardinality exactly two, they form edges in a graph whose vertices are the universal set. A system of subsets with no restrictions on the cardinality of its members is called a hypergraph. It often can be profitable to consider whether your problem has a graph-theoretical analogy, like connected components or shortest path in a hypergraph. Your primary alternatives for representing arbitrary systems of subsets are: ●

●

Bit vectors - If your universal set U contains n items, an n-bit vector or array can represent any . Bit i will be 1 if , otherwise bit i is 0. Since only one bit is used per element, subset bit vectors can be very space efficient for surprisingly large values of |U|. Element insertion and deletion simply flips the appropriate bit. Intersection and union are done by ``and-ing'' or ``or-ing'' the bits together. The only real drawback of a bit vector is that for sparse subsets, it takes O(n) time to explicitly identify all members of S. Containers or dictionaries - A subset can also be represented using a linked list, array, binary tree, or dictionary containing exactly the elements in the subset. No notion of a fixed universal set is needed for such a data structure. For sparse subsets, dictionaries can be more space and time efficient than bit vectors and easier to work with and program. For efficient union and intersection operations, it pays to keep the elements in each subset sorted, so a linear-time traversal through both subsets identifies all duplicates.

In many applications, the subsets are all pairwise disjoint, meaning that each element is in exactly one subset. For example, consider maintaining the connected components of a graph or the party affiliations of politicians. Each vertex/hack is in exactly one component/party. Such a system of subsets is called a set partition. Algorithms for constructing partitions of a given set are provided in Section

.

For data structures, the primary issue is maintaining a given set partition as things change over time, perhaps as edges are added or party members defect. The queries we are interested in include ``which set is a particular item in?'' and ``are two items in the same set?'' as we modify the set by (1) changing one item, (2) merging or unioning two sets, or (3) breaking a set apart. Your primary options are: ●

●

Dictionary with subset attribute - If each item in a binary tree has associated a field recording the name of the subset it is in, set identification queries and single element modifications can be performed in the time it takes to search in the dictionary, typically . However, operations like performing the union of two subsets take time proportional to (at least) the sizes of the subsets, since each element must have its name changed. The need to perform such union operations quickly is the motivation for the ... Union-Find Data Structure - Suppose we represent a subset using a rooted tree, where each node points to its parent instead of its children. Further, let the name of the subset be the name of the item that is the root. Finding out which subset we are in is simple, for we keep traversing up the

file:///E|/BOOK/BOOK3/NODE133.HTM (2 of 4) [19/1/2003 1:30:10]

Set Data Structures

parent pointers until we hit the root. Unioning two subsets is also easy. Just make the root of one of two trees point to the other, so now all elements have the same root and thus the same subset name. Certain details remain, such as which subset should be the ultimate root of a union, but these are described in most every algorithms text. Union-Find is a fast, extremely simple data structure that every programmer should know about. It does not support breaking up subsets created by unions, but usually this is not an issue. Neither of these options provides access to all of the items in a particular subset without traversing all the items in the set. However, both can be appropriately augmented with extra pointers if it is important that this operation be fast. Implementations: LEDA (see Section ) provides dictionary data structures to maintain sets and the union-find data structure to maintain set partitions, all in C++. LINK is an environment for combinatorial computing that provides special support for hypergraphs, including visualization of hypergraphs. Although written in C++, it provides an additional Scheme language interface for interacting with the graphs. LINK is available from http://dimacs.rutgers.edu/Projects/LINK.html. Many textbooks contain implementations of the union-find data structure, including [MS91] (see Section ). An implementation of union-find underlies any implementation of Kruskal's minimum spanning tree algorithm. Section

contains a selection of minimum spanning tree codes.

Notes: Optimal algorithms for such set operations as intersection and union were presented in [Rei72]. Good expositions on set data structures include [AHU83]. Galil and Italiano [GI91] survey data structures for disjoint set union. Expositions on the union-find data structure appear in most algorithm texts, including [CLR90, MS91]. The upper bound of on m union-find operations on an n-element set is due to Tarjan [Tar75], as is a matching lower bound on a restricted model of computation [Tar79]. The inverse Ackerman function grows notoriously slowly, so this performance is close to linear. Expositions on the Ackerman bound include [AHU74]. An interesting connection between the worst-case of union-find and the length of Davenport-Schintzl sequences, a combinatorial structure that arises in computational geometry, is established in [SA95]. The power set of a set S is the collection of all subsets of S. Explicit manipulation of power sets quickly becomes intractable due to their size. Implicit representations of power sets in symbolic form becomes necessary for nontrivial computations. See [BCGR92] for algorithms for and computational file:///E|/BOOK/BOOK3/NODE133.HTM (3 of 4) [19/1/2003 1:30:10]

Set Data Structures

experience with symbolic power set representations. Related Problems: Generating subsets (see page page

), minimum spanning tree (see page

), generating partitions (see page

).

Next: Kd-Trees Up: Data Structures Previous: Graph Data Structures Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE133.HTM (4 of 4) [19/1/2003 1:30:10]

), set cover (see

Kd-Trees

Next: Numerical Problems Up: Data Structures Previous: Set Data Structures

Kd-Trees

Input description: A set S of n points in k dimensions. Problem description: Construct a tree that partitions the space by half-planes such that each point is contained in its own box-shaped region. Discussion: Although many different flavors of kd-trees have been devised, their purpose is always to hierarchically decompose space into a relatively small number of cells such that no cell contains too many input objects. This provides a fast way to access any input object by position. We traverse down the hierarchy until we find the cell containing the object and then scan through the few objects in the cell to identify the right one. Typical algorithms construct kd-trees by partitioning point sets. Each node in the tree is defined by a plane through one of the dimensions that partitions the set of points into left/right (or up/down) sets, each with half the points of the parent node. These children are again partitioned into equal halves, using planes through a different dimension. Partitioning stops after levels, with each point in its own leaf cell. Alternate kd-tree construction algorithms insert points incrementally and divide the appropriate cell, although such trees can become seriously unbalanced.

file:///E|/BOOK/BOOK3/NODE134.HTM (1 of 4) [19/1/2003 1:30:12]

Kd-Trees

The cutting planes along any path from the root to another node defines a unique box-shaped region of space, and each subsequent plane cuts this box into two boxes. Each box-shaped region is defined by 2k planes, where k is the number of dimensions. Indeed, the `kd' in kd-tree is short for k-dimensional tree. In any search performed using a kd-tree, we maintain the current region defined by the intersection of these half-spaces as we move down the tree. Different flavors of kd-trees differ in exactly how the splitting plane is selected. Options include: ● ●

●

Cycling through the dimensions - partition first on , then before cycling back to . Cut along the largest dimension - select the partition dimension so as to make the resulting boxes as square or cube-like as possible. Selecting a plane to partition the points in half does not mean selecting a splitter in the middle of the box-shaped regions, since all the points may be in the left side of the box. Quadtrees or Octtrees - Instead of partitioning with single planes, use all axis-parallel planes that pass through a given partition point. In two dimensions, this means creating four child cells, in 3D this means eight child cells. Quadtrees seem particularly popular on image data, where leaf cells imply that all pixels in the regions have the same color.

Nonorthogonal (i.e. not axis-parallel) cutting planes have also been used, although they make maintaining the cell boundaries more complicated. Ideally, our partitions evenly split both the space (ensuring nice, fat, regular regions) and the set of points (ensuring a log height tree) evenly, but this can be impossible for a given point set. The advantages of fat cells become clear in many applications of kd-trees: ●

●

●

●

Point location - To identify which cell a query point q lies in, we start at the root and test which side of the partition plane contains q. By repeating this process on the appropriate child node, we travel the tree to find the leaf cell containing q in time proportional to its height. See Section for more on point location. Nearest neighbor search - To find the point in S closest to a query point q, we perform point location to find the cell c containing q. Since c is bordered by some point p, we can compute the distance d(p,q) from p to q. Point p is likely very close to q, but it might not be the single closest neighbor. Why? Suppose q lies right at the boundary of a cell. Then q's nearest neighbor might lie just to the left of the boundary in another cell. Thus we must traverse all cells that lie within a distance of d(p,q) of cell c and verify that none of them contain closer points. With nice, fat cells, for more on nearest neighbor search. very few cells should need to be tested. See Section Range search - Which points lie within a query box or region? Starting from the root, check to see whether the query region intersects or contains the cell defining the current node. If it does, check the children; if not, none of the leaf cells below this node can possibly be of interest. We for more on range search. quickly prune away the irrelevant portions of the space. See Section Partial key search - Suppose we want to find a point p in S, but we do not have full information

file:///E|/BOOK/BOOK3/NODE134.HTM (2 of 4) [19/1/2003 1:30:12]

Kd-Trees

about p. Say we are looking for someone of age 35 and height 5'8'' but of unknown weight in a 3dtree with dimensions age, weight, and height. Starting from the root, we can identify the correct decendant for all but the weight dimension. To be sure we find the right point, we must search both children of this node. We are better off the more fields we know, but such partial key search can be substantially faster than checking all points against the key. Kd-trees are most effective data structures for small and moderate numbers of dimensions, say from 2 up to maybe 20 dimensions. As the dimensionality increases, they lose effectiveness, primarily because the ratio of the volume of a unit sphere in k-dimensions shrinks exponentially compared to a unit cube in kdimensions. Thus exponentially many cells will have to be searched within a given radius of a query point, say for nearest-neighbor search. Also, the number of neighbors for any cell grows to 2k and eventually become unmanageable. The bottom line is that you should try to avoid working in high-dimensional spaces, perhaps by discarding the least important dimensions. Implementations: Ranger is a tool for visualizing and experimenting with nearest neighbor and orthogonal range queries in high-dimensional data sets, using multidimensional search trees. Four different search data structures are supported by Ranger: naive kd-trees, median kd-trees, nonorthogonal kd-trees, and the vantage point tree. For each of these, Ranger supports queries in up to 25 dimensions under any Minkowski metric. It includes generators for a variety of point distributions in arbitrary dimensions. Finally, Ranger provides a number of features to aid in visualizing multidimensional data, best illustrated by the accompanying video [MS93]. To identify the most appropriate projection at a matrix of all two-dimensional projections of the data set. Ranger is glance, Ranger provides a written in C, runs on Silicon Graphics and HP workstations, and is available from the algorithm repository. The 1996 DIMACS implementation challenge focuses on data structures for higher-dimensional data sets. The world's best kd-tree implementations were likely to be identified in the course of the challenge, and they are accessible from http://dimacs.rutgers.edu/. Bare bones implementations in C of kd-tree and quadtree data structures appear in [GBY91]. See Section for details on how to ftp them. Notes: The best reference on kd-trees and other spatial data structures are two volumes by Samet [Sam90a, Sam90b], in which all major variants are developed in substantial detail. Bentley [Ben75] is generally credited with developing kd-trees, although they have the typically murky history associated with most folk data structures. The most reliable history is likely from Samet [Sam90b].

file:///E|/BOOK/BOOK3/NODE134.HTM (3 of 4) [19/1/2003 1:30:12]

Kd-Trees

An exposition on kd-trees for orthogonal range queries in two dimensions appears in [PS85]. Expositions of grid files and other spatial data structures include [NH93]. Algorithms that quickly produce a point provably close to the query point are a recent development in higher-dimensional nearest neighbor search. A sparse weighted-graph structure is built from the data set, and the nearest neighbor is found by starting at a random point and walking greedily in the graph towards the query point. The closest point found during several random trials is declared the winner. Similar data structures hold promise for other problems in high-dimensional spaces. See [AM93, AMN 94]. Related Problems: Nearest-neighbor search (see page (see page

), point location (see page

).

Next: Numerical Problems Up: Data Structures Previous: Set Data Structures Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE134.HTM (4 of 4) [19/1/2003 1:30:12]

), range search

Numerical Problems

Next: Solving Linear Equations Up: A Catalog of Algorithmic Previous: Kd-Trees

Numerical Problems If most problems you encounter are numerical in nature, there is a good chance that you are reading the wrong book. Numerical Recipes [PFTV86] gives a terrific overview to the fundamental problems in numerical computing, including linear algebra, numerical integration, statistics, and differential equations. Different flavors of the book include source code for all the algorithms in C, Pascal, and Fortran. Their coverage is skimpier on the combinatorial/numerical problems we consider in this section, but you should be aware of that book. Numerical algorithms tend to be different beasts than combinatorial algorithms, for at least two distinct reasons: ●

●

Issues of Precision and Error - Numerical algorithms typically perform repeated floating-point computations, which accumulate error at each operation until, eventually, the results are meaningless. An amusing example [SK93] concerns the Vancouver Stock Exchange, which over a twenty-two month period accumulated sufficient round-off error to reduce its index from the correct value of 1098.982 to 574.081. A simple and dependable way to test for round-off errors in numerical programs is to run them both at single and double precision, and then think hard whenever there is a disagreement. Extensive Libraries of Codes Large, high-quality libraries of numerical routines have existed since the 1960s, which is still not the case for combinatorial algorithms. This is true for several reasons, including (1) the early emergence of Fortran as a standard for numerical computation, (2) the nature of numerical computations to remain independent rather than be embedded within large applications, and (3) the existence of large scientific communities needing general numerical libraries. Regardless of why, you should exploit this software base. There is probably no reason to implement algorithms for any of the problems in this section instead of stealing existing codes. Searching Netlib (see Section

) is always a good place to start.

Most scientist's and engineer's ideas about algorithms derive from Fortran programming and numerical methods, while computer scientists grew up programming with pointers and recursion, and so are comfortable with the more sophisticated data structures required for combinatorial algorithms. Both sides can and should learn from each other, since several problems such as pattern recognition can be modeled file:///E|/BOOK/BOOK3/NODE135.HTM (1 of 3) [19/1/2003 1:30:13]

Numerical Problems

either numerically or combinatorially. There is a vast literature on numerical algorithms. In addition to Numerical Recipes, recommended books include: ●

●

●

●

● ● ● ● ● ● ● ● ● ● ●

Skeel and Keiper [SK93] - A readable and interesting treatment of basic numerical methods, avoiding overly detailed algorithm descriptions through its use of the computer algebra system Mathematica. I like it. Pizer and Wallace [PW83] - A numerical analysis book written for computer scientists, not engineers. The organization is by issues instead of problems. A different but interesting perspective. Cheney and Kincaid [CK80] - A traditional Fortran-based numerical analysis text, with discussions of optimization and Monte Carlo methods in addition to such standard topics as rootfinding, numerical integration, linear systems, splines, and differential equations. Buchanan and Turner [BT92] - Thorough language-independent treatment of all standard topics, including parallel algorithms. Most comprehensive of the texts described here.

Solving Linear Equations Bandwidth Reduction Matrix Multiplication Determinants and Permanents Constrained and Unconstrained Optimization Linear Programming Random Number Generation Factoring and Primality Testing Arbitrary-Precision Arithmetic Knapsack Problem Discrete Fourier Transform

Next: Solving Linear Equations Up: A Catalog of Algorithmic Previous: Kd-Trees Algorithms file:///E|/BOOK/BOOK3/NODE135.HTM (2 of 3) [19/1/2003 1:30:13]

Numerical Problems

Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE135.HTM (3 of 3) [19/1/2003 1:30:13]

Solving Linear Equations

Next: Bandwidth Reduction Up: Numerical Problems Previous: Numerical Problems

Solving Linear Equations

Input description: An on n variables.

matrix A, and an

Problem description: What is the vector x such that

vector b, together representing m linear equations

?

Discussion: The need to solve linear systems arises in an estimated 75% of all scientific computing problems [DB74]. For example, applying Kirchhoff's laws to analyze electric circuits generates a system of equations, the solution of which gives currents through each branch of the circuit. Analysis of the forces acting on a mechanical truss generates a similar set of equations. Even finding the point of intersection between two or more lines reduces to solving a (small) linear system. Not all systems of equations have solutions; consider the equations 2x+3y = 5 and 2x+3y = 6. Some systems of equations have multiple solutions; consider the equations 2x+3y=5 and 4x+6y=10. Such degenerate systems of equations are called singular, and they can be recognized by testing whether the determinant of the coefficient matrix is zero. Solving linear systems is a problem of such scientific and commercial importance that excellent codes are readily available. There is likely no good reason to implement your own solver, even though the basic file:///E|/BOOK/BOOK3/NODE136.HTM (1 of 4) [19/1/2003 1:30:15]

Solving Linear Equations

algorithm (Gaussian elimination) is one we learned in high school. working with large systems.

This is especially true if you are

Gaussian elimination is based on the fact that the solution to a system of linear equations is invariant under scaling (multiplying both sides by a constant; i.e. if x=y, then 2x=2y) and adding equations (i.e. the solution to the equations x=y and w=z is the same as the solution to x=y and x+w=y+z). Gaussian elimination scales and adds equations so as to eliminate each variable from all but one equation, leaving the system in such a state that the solution can just be read off from the equations. The time complexity of Gaussian elimination on an system of equations is , since for the ith variable we add a scaled copy of the n-term ith row to each of the n-1 other equations. On this problem, however, constants matter. Algorithms that only partially reduce the coefficient matrix and then backsubstitute to get the answer use 50% fewer floating-point operations than the naive algorithm. Issues to worry about include: ●

Are roundoff errors and numerical stability affecting my solution? - Implementing Gaussian elimination would be quite straightforward except for round-off errors, which accumulate with each row operation and can quickly wreak havoc on the solution, particularly with matrices that are almost singular. To eliminate the danger of numerical errors, it pays to substitute the solution back into each of the original equations and test how close they are to the desired value. Iterative methods for solving linear systems refine initial solutions to obtain more accurate answers - good linear systems packages will include such routines.

●

●

The key to minimizing roundoff errors in Gaussian elimination is selecting the right equations and variables to pivot on, and to scale the equations so as to eliminate large coefficients. This is an art as much as a science, which is why you should use one of the many well-crafted library routines described below. Which routine in the library should I use? - Selecting the right code is also somewhat of an art. If you are taking your advice from this book, you should start with the general linear system solvers. Hopefully they will suffice for your needs. But search through the manual for more efficient procedures for solving special types of linear systems. If your matrix happens to be one of these special types, the solution time can reduce from cubic to quadratic or even linear. Is my system sparse? - The key to recognizing that you might have a special-case linear system is establishing how many matrix elements you really need to describe A. If there are only a few nonzero elements, your matrix is sparse and you are in luck. If these few nonzero elements are clustered near the diagonal, your matrix is banded and you are in even more luck. Algorithms for . Many other regular patterns of reducing the bandwidth of a matrix are discussed in Section sparse matrices can also be exploited, so see the manual of your solver or a better book on numerical analysis for details.

file:///E|/BOOK/BOOK3/NODE136.HTM (2 of 4) [19/1/2003 1:30:15]

Solving Linear Equations ●

Will I be solving many systems with the same coefficient matrix? - In certain applications, such as least-squares curve fitting and differential equations, we have to solve repeatedly with different b vectors. For efficiency, we seek to preprocess A to make this easier. The lowerupper or LU-decomposition of A creates lower- and upper-triangular matrices L and U such that . We can use this decomposition to solve , since

This is efficient since backsubstitution solves a triangular system of equations in quadratic time. Solving and then gives the solution x using two steps instead of one time. step, once the LU-decomposition has been found in The problem of solving linear systems is equivalent to that of matrix inversion, since , where is the identity matrix. However, avoid it, since matrix inversion proves to be three times slower than Gaussian elimination. LU-decompositions prove useful in inverting matrices as well as computing determinants (see Section

).

Implementations: The library of choice for solving linear systems is apparently LAPACK, a descendant of LINPACK [DMBS79]. Both of these Fortran codes, as well as many others, are available from Netlib. See Section

.

Algorithm 533 [She78], Algorithm 576 [BS81], and Algorithm 578 [DNRT81] of the Collected Algorithms of the ACM are Fortran codes for Gaussian elimination. Algorithm 533 is designed for sparse systems, algorithm 576 to minimize roundoff errors, and algorithm 578 to optimize virtual memory performance. Algorithm 645 [NW86] is a Fortran code for testing matrix inversion programs. See Section

for details on fetching these programs.

Numerical Recipes [PFTV86] provides routines for solving linear systems. However, there is no compelling reason to use these ahead of the free codes described in this section. C++ implementations of LEDA (see Section

algorithms to solve linear equations and invert matrices are embedded in

).

Notes: Good expositions on algorithms for Gaussian elimination and LU-decomposition include [CLR90] and a host of numerical analysis texts [BT92, CK80, SK93]. Parallel algorithms for linear systems are discussed in [Ort88]. Solving linear systems is one of relatively few problems where parallel architectures are widely used in practice.

file:///E|/BOOK/BOOK3/NODE136.HTM (3 of 4) [19/1/2003 1:30:15]

Solving Linear Equations

Matrix inversion, and hence linear systems solving, can be done in matrix multiplication time using Strassen's algorithm plus a reduction. Good expositions on the equivalence of these problems include [AHU74, CLR90]. Certain types of nonsparse systems can be solved efficiently via special algorithms. In particular, Toeplitz matrices are constructed so that all the elements along a particular diagonal are identical, and Vandermonde matrices are defined by an n-element vector x where . Related Problems: Matrix multiplication (see page

), determinant/permanent (see page

Next: Bandwidth Reduction Up: Numerical Problems Previous: Numerical Problems Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE136.HTM (4 of 4) [19/1/2003 1:30:15]

).

Bandwidth Reduction

Next: Matrix Multiplication Up: Numerical Problems Previous: Solving Linear Equations

Bandwidth Reduction

Input description: A graph G=(V,E), representing an

matrix M of zero and nonzero elements.

Problem description: Which permutation p of the vertices of V minimizes the length of the longest edge when the vertices are ordered on a line, i.e. minimizes ? Discussion: Bandwidth reduction lurks as a hidden but important problem for both graphs and matrices, and it is important to see how it arises so as to properly recognize it. Applied to matrices, it permutes the rows and columns of a sparse matrix so as to minimize the distance b of any nonzero entry from the center diagonal. This is important in solving linear systems, because Gaussian elimination (see Section ) can be performed in algorithm if b << n.

on matrices of bandwidth b. This is a big win over the general

Bandwidth minimization on graphs arises in more subtle ways. Arranging a set of n circuit components in a line on a circuit board so as to minimize the length of the longest wire (and hence time delay) is a bandwidth problem, where each vertex of our graph corresponds to a circuit component and there is an edge for every wire linking two components. Alternatively, consider a hypertext application where we must store large objects (say images) on a magnetic tape. From each image there is a set of possible images we can go to next (i.e. the hyperlinks). To minimize the search time, we seek to place linked file:///E|/BOOK/BOOK3/NODE137.HTM (1 of 3) [19/1/2003 1:30:16]

Bandwidth Reduction

images near each other on the tape. This is exactly the bandwidth problem. More general formulations, such as rectangular circuit layouts and magnetic disks, inherit the same hardness and classes of heuristics from the linear versions. Unfortunately, bandwidth minimization is NP-complete. It stays NP-complete even if the input graph is a tree whose maximum vertex degree is 3, which is an unusually strong condition. Further, there is no known approximation algorithm for bandwidth reduction, even for trees. Thus our only options are brute-force search or ad hoc heuristics. Fortunately, ad hoc heuristics have been well-studied in the numerical analysis community, and production-quality implementations of the best heuristics are available. These are based on performing a breadth-first search from a given vertex v, where v is placed at the leftmost point of the ordering. All of the vertices that are distance 1 from v are placed to its immediate right, followed by all the vertices at distance 2, and so forth until we reach the vertex furthest from v. We then continue the breadth-first search from the vertex immediately to the right of v until all vertices in G are accounted for. The popular heuristics differ according to how many different start vertices are considered and how equidistant vertices are ordered among themselves. However, breaking ties with low-degree vertices over to the left seems to be a good idea. Implementations of the most popular heuristics, the Cuthill-McKee and Gibbs-Poole-Stockmeyer algorithms, are discussed in the implementation section. The worst case of the Gibbs-Poole-Stockmeyer algorithm is , which would wash out any possible savings in solving linear systems, but its performance in practice is close to linear. Brute-force search programs can find the exact minimum bandwidth work by backtracking through the set of n! possible permutations of vertices. Considerable pruning can be achieved to reduce the search space by starting with a good heuristic bandwidth solution and alternately adding vertices to the left- and rightmost open slots in the partial permutation. The first edge connecting a vertex on the left to a vertex on the right will likely define an edge whose length is greater than our best example to date, thus leading to fast pruning. In our experience, graphs of size n=30 or so can be solved to optimality. See the . However, for almost any application such an exact solution discussion on backtracking in Section will not be worth the expense of finding it. Implementations: Fortran language implementations of both the Cuthill-McKee algorithm [CGPS76, Gib76, CM69] and the Gibbs-Poole-Stockmeyer algorithm [Lew82, GPS76] are available from Netlib. See Section . Empirical evaluations of these and other algorithms on a test suite of 30 matrices are discussed in [Eve79b], showing Gibbs-Poole-Stockmeyer to be the consistent winner. Brute-force implementations written by Ingmar Bitter, Christian Joita, and Dario Vlah in C and C++ as Stony Brook class projects and capable of solving instances of size n=30 to optimality are provided on the algorithm repository http://www.cs.sunysb.edu/ algorith. file:///E|/BOOK/BOOK3/NODE137.HTM (2 of 3) [19/1/2003 1:30:16]

Bandwidth Reduction

Notes: An excellent survey on graph-theoretic and algorithmic results on the bandwidth problem to 1981 appears in [CCDG82]. Ad hoc heuristics have been widely studied, a tribute to its importance in numerical computation. Everstine [Eve79b] cites no less than 49 different bandwidth reduction algorithms! The hardness of the bandwidth problem was first established by Papadimitriou [Pap76b], and its hardness on trees of maximum degree 3 in [GGJK78]. There are algorithms that run in polynomial time for fixed bandwidth k [Sax80]. An exposition on the hardness of the linear arrangement problem appears in [Eve79a]. Related Problems: Solving linear equations (see page

), topological sorting (see page

Next: Matrix Multiplication Up: Numerical Problems Previous: Solving Linear Equations Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE137.HTM (3 of 3) [19/1/2003 1:30:16]

).

Matrix Multiplication

Next: Determinants and Permanents Up: Numerical Problems Previous: Bandwidth Reduction

Matrix Multiplication

Input description: An Problem description: The

matrix A and a matrix

matrix B. .

Discussion: Although matrix multiplication is an important problem in linear algebra, its main significance for combinatorial algorithms is its equivalence to a variety of other problems, such as transitive closure and reduction, solving linear systems, and matrix inversion. Thus a faster algorithm for matrix multiplication implies faster algorithms for all of these problems. Matrix multiplication arises in its own right in computing the results of such coordinate transformations as scaling, rotation, and translation for robotics and computer graphics. The straightforward algorithm to compute the product of z) time and is tough to beat in practice:

for i=1 to x do

file:///E|/BOOK/BOOK3/NODE138.HTM (1 of 4) [19/1/2003 1:30:18]

matrix A and

matrix B runs in O(x y

Matrix Multiplication

for j = 1 to z

In two multiplying bandwidth-b matrices, where all nonzero elements of A and B lie within b elements of the main diagonals, a speedup to O(x b z) is possible, since zero elements will not contribute to the product. Asymptotically faster algorithms for matrix multiplication exist, based on clever divide-and-conquer recurrences. However, these prove difficult to program and require very large matrices to beat the trivial algorithm is unlikely to beat algorithm. In particular, some empirical results show that Strassen's , and it is less numerically stable to boot. Other studies have the straightforward algorithm for . Still, I consider it unlikely been more encouraging, claiming that the crossover point is as low as that you will speed up any serious application by implementing Strassen's algorithm. There is a better way to save computation when you are multiplying a chain of more than two matrices together. Recall that multiplying an matrix by a matrix creates an matrix. Thus multiplying a chain of matrices from left to right might create large intermediate matrices, each taking a lot of time to compute. Matrix multiplication is not commutative, but it is associative, so we can parenthesize the chain in whatever manner we deem best without changing the final product. A standard dynamic programming algorithm can be used to construct the optimal parenthesization. Whether it pays to do this optimization will depend upon whether your matrices are large enough or your chain is multiplied often enough to justify it. Note that we are optimizing over the sizes of the dimensions in the chain, not the actual matrices themselves. If all your matrices are the same dimensions, you are out of luck, since no such optimization is possible. Matrix multiplication has a particularly interesting interpretation in counting the number of paths between two vertices in a graph. Let A be the adjacency matrix of a graph G, meaning A[i,j] = 1 if there is an edge between i and j. Otherwise, A[i,j] = 0. Now consider the square of this matrix, . If , this means that there must be a k such that A[i,k]=A[k,j]=1, so i to k to j is a counts the number of paths of length exactly k between i path of length 2 in G. More generally, and j. This count includes nonsimple paths, where vertices are repeated, such as i to k to i. Implementations: The quick and dirty algorithm will be your best bet unless your matrices are very large. For example, [CLR90] suggests that n>45 before you have a hope of winning. Experimental results suggest that n > 100 is more realistic [CR76], with Bailey [BLS91] finding a crossover point of n=128 for Cray systems. Strassen's algorithm is difficult to implement efficiently because of the data structures required to maintain the array partitions. That said, an implementation of Strassen's algorithm in Mathematica by Stan Wagon is offered ``without promise of efficiency'' on the algorithm repository WWW site. file:///E|/BOOK/BOOK3/NODE138.HTM (2 of 4) [19/1/2003 1:30:18]

Matrix Multiplication

The linear algebra library of choice is LAPACK, a descendant of LINPACK [DMBS79], which includes several routines for matrix multiplication. These Fortran codes are available from Netlib as discussed in Section

.

Algorithm 601 [McN83] of the Collected Algorithms of the ACM is a sparse matrix package written in Fortran that includes routines to multiply any combination of sparse and dense matrices. See Section for details. ) is an algorithm animation system for UNIX and X-windows that includes an XTango (see Section matrix multiplication algorithm. A C++, implementation of matrix animation of the multiplication is embedded in LEDA (see Section

).

Notes: Winograd's algorithm for fast matrix multiplication reduces the number of multiplications by a factor of two over the straightforward algorithm. It is implementable, although the additional bookkeeping required makes it doubtful whether it is a win. Expositions on Winograd's algorithm [Win68] include [CLR90, Man89, Win80]. In my opinion, the history of theoretical algorithm design began when Strassen published his time matrix multiplication algorithm. For the first time, improving an algorithm in the asymptotic sense became a respected goal in its own right. Good expositions on Strassen's algorithm [Str69] include [Baa88, CLR90, Cra94]. Progressive improvements to Strassen's algorithm have gotten progressively less practical. The current best result for matrix multiplication is Coppersmith and Winograd's [CW87] algorithm, while the conjecture is that suffices. The interest in the squares of graphs goes beyond counting paths. Fleischner [Fle74] proved that the square of any biconnected graph has a Hamiltonian cycle. See [LS95] for results on finding the square roots of graphs, i.e. finding A given . The problem of Boolean matrix multiplication can be reduced to that of general matrix multiplication [CLR90]. The four-Russians algorithm for Boolean matrix multiplication [ADKF70] uses preprocessing to construct all subsets of rows for fast retreival in performing the actual multiplication, yielding a . Additional preprocessing can improve this to [Ryt85]. An complexity of exposition on the four-Russians algorithm, including this speedup, appears in [Man89]. Good expositions of the matrix-chain algorithm include [Baa88, CLR90], where it is a standard example of dynamic programming.

file:///E|/BOOK/BOOK3/NODE138.HTM (3 of 4) [19/1/2003 1:30:18]

Matrix Multiplication

Related Problems: Solving linear equations (see page

), shortest path (see page

).

Next: Determinants and Permanents Up: Numerical Problems Previous: Bandwidth Reduction Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE138.HTM (4 of 4) [19/1/2003 1:30:18]

Determinants and Permanents

Next: Constrained and Unconstrained Optimization Up: Numerical Problems Previous: Matrix Multiplication

Determinants and Permanents

Input description: An

matrix M.

Problem description: What is the determinant |M| or the permanent perm(M) of the matrix m? Discussion: Determinants of matrices provide a clean and useful abstraction in linear algebra that can used to solve a variety of problems:

●

Testing whether a matrix is singular, meaning that the matrix does not have an inverse. A matrix M is singular iff |M| = 0. Testing whether a set of d points lies on a plane in fewer than d dimensions. If so, the system of equations they define is singular, so |M| = 0. Testing whether a point lies to the left or right of a line or plane. This problem reduces to

●

. testing whether the sign of a determinant is positive or negative, as discussed in Section Computing the area or volume of a triangle, tetrahedron, or other simplicial complex. These

●

●

quantities are a function of the magnitude of the determinant, as discussed in Section

file:///E|/BOOK/BOOK3/NODE139.HTM (1 of 3) [19/1/2003 1:30:19]

.

Determinants and Permanents

The determinant of a matrix M is defined as the sum over all n! possible permutations of M:

where

of the n columns

is the number of pairs of elements out of order (called inversions) in permutation

.

A direct implementation of this definition yields an O(n!) algorithm, as does the cofactor expansion method we learned in high school. However, better algorithms are available to evaluate determinants . The determinant of M is simply the based on LU-decomposition. They are discussed in Section time. product of the diagonal elements of the LU-decomposition of M, which can be found in A closely related function, called the permanent, arises often in combinatorial problems. For example, the permanent of the adjacency matrix of a graph G counts the number of perfect matchings in G. The permanent of a matrix M is defined by

differing from the determinant only in that all products are positive. Surprisingly, it is NP-hard to compute the permanent, even though the determinant can easily be computed in time. The fundamental difference is that , while . Fortunately, there are permanent algorithms that prove to be considerably faster than the O(n!) definition, running in time. Thus finding the permanent of a matrix is not out of the realm of possibility. Implementations: The linear algebra package LINPACK contains a variety of Fortran routines for computing determinants, optimized for different data types and matrix structures. It can be obtained from Netlib, as discussed in Section in LEDA (see Section

. A C++ program to compute determinants in

time is embedded

).

Nijenhuis and Wilf [NW78] provide an efficient Fortran routine to compute the permanent of a matrix. See Section

.

Notes: Cramer's rule reduces the problems of matrix inversion and solving linear systems to that of computing determinants. However, algorithms based on LU-determination are faster. See [BM53] for an

file:///E|/BOOK/BOOK3/NODE139.HTM (2 of 3) [19/1/2003 1:30:19]

Determinants and Permanents

exposition on Cramer's rule. Determinants can be computed in

time using fast matrix multiplication, as shown in [AHU83].

Section discusses such algorithms. A fast algorithm for computing the sign of the determinant, important problem for performing robust geometric computations, is due to Clarkson [Cla92].

an

The problem of computing the permanent was shown to be #P-complete by Valiant [Val79], where #P is the class of problems solvable on a ``counting'' machine in polynomial time. A counting machine returns the number of distinct solutions to a problem. Counting the number of Hamiltonian cycles in a graph is a #P-complete problem that is trivially NP-hard (and presumably harder), since any count greater than zero proves that the graph is Hamiltonian. Counting problems can be #P-complete even if the corresponding decision problem can be solved in polynomial time, as shown by the permanent and perfect matchings. Minc [Min78] is the definitive work on permanents. A variant of an -time algorithm due to Ryser for computing the permanent is presented in [NW78]. Recently, probabilistic algorithms have been developed for estimating the permanent [FJ95]. Related Problems: Solving linear systems (see page (see page

), matching (see page

), geometric primitives

).

Next: Constrained and Unconstrained Optimization Up: Numerical Problems Previous: Matrix Multiplication Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE139.HTM (3 of 3) [19/1/2003 1:30:19]

Constrained and Unconstrained Optimization

Next: Linear Programming Up: Numerical Problems Previous: Determinants and Permanents

Constrained and Unconstrained Optimization

Input description: A function

.

Problem description: What point

maximizes (or minimizes) the function f?

Discussion: Most of this book concerns algorithms that optimize one thing or another. This section considers the general problem of optimizing functions where, due to lack of structure or knowledge, we are unable to exploit the problem-specific algorithms seen elsewhere in this book. Optimization arises whenever there is an objective function that must be tuned for optimal performance. Suppose we are building a program to identify good stocks to invest in. We have available certain financial data to analyze, such as the price-earnings ratio, the interest and inflation rates, and the stock price, all as a function of time t. The key question is how much weight we should give to each of these factors, where these weights correspond to coefficents of a formula:

We seek the numerical values , , , whose stock-goodness function does the best job of evaluating stocks. Similar issues arise in tuning evaluation functions for game playing programs such as chess. file:///E|/BOOK/BOOK3/NODE140.HTM (1 of 4) [19/1/2003 1:30:21]

Constrained and Unconstrained Optimization

Unconstrained optimization problems also arise in scientific computation. Physical systems from protein structures to particles naturally seek to minimize their ``energy functions.'' Thus programs that attempt to simulate nature often define energy potential functions for the possible configurations of objects and then take as the ultimate configuration the one that minimizes this potential. Global optimization problems tend to be hard, and there are lots of ways to go about them. Ask the following questions to steer yourself in the right direction: ●

●

●

●

Am I doing constrained or unconstrained optimization? - In unconstrained optimization, there are no limitations on the values of the parameters other than that they maximize the value of f. Often, however, there are costs or constraints on these parameters. These constraints make certain points illegal, points that might otherwise be the global optimum. Constrained optimization problems typically require mathematical programming approaches like linear programming, . discussed in Section Is the function I am trying to optimize described by a formula or data? - If the function that you seek to optimize is presented as an algebraic formula (such as the minimum of ), the solution is to analytically take its derivative f'(n) and see for which points p' we have f'(p') = 0. These points are either local maxima or minima, which can be distinguished by taking a second derivative or just plugging back into f and seeing what happens. Symbolic computation systems such as Mathematica and Maple are fairly effective at computing such derivatives, although using computer algebra systems effectively is somewhat of a black art. They are definitely worth a try, however, and you can always use them to plot a picture of your function to get a better idea of what you are dealing with. How expensive is it to compute the function at a given point? - If the function f is not presented as a formula, what to do depends upon what is given. Typically, we have a program or subroutine that evaluates f at a given point, and so can request the value of any given point on demand. By calling this function, we can poke around and try to guess the maxima. Our freedom to search in such a situation depends upon how efficiently we can evaluate f. If f is just a complicated formula, evaluation will be very fast. But suppose that f represents the effect of the coefficients on the performance of the board evaluation function in a computer chess program, such that is how much a pawn is worth, is how much a bishop is worth, and so forth. To evaluate a set of coefficients as a board evaluator, we must play a bunch of games with it or test it on a library of known positions. Clearly, this is time-consuming, so we must be frugal in the number of evaluations of f we use. How many dimensions do we have? How many do we need? - The difficulty in finding a global maximum increases rapidly with the number of dimensions (or parameters). For this reason, it often pays to reduce the dimension by ignoring some of the parameters. This runs counter to intuition, for the naive programmer is likely to incorporate as many variables as possible into their evaluation function. It is just too hard to tweak such a complicated function. Much better is to start with the 3 to 5 seemingly most important variables and do a good job optimizing the coefficients for these.

file:///E|/BOOK/BOOK3/NODE140.HTM (2 of 4) [19/1/2003 1:30:21]

Constrained and Unconstrained Optimization ●

How smooth is my function? The main difficulty of global optimization is getting trapped in local optima. Consider the problem of finding the highest point in a mountain range. If there is only one mountain and it is nicely shaped, we can find the top by just walking in whatever direction is up. However, if there are many false summits or other mountains in the area, it is difficult to convince ourselves whether we are really at the highest point. Smoothness is the property that enables us to quickly find the local optimum from a given point. We assume smoothness in seeking the peak of the mountain by walking up. If the height at any given point was a completely random function, there would be no way we could find the optimum height short of sampling every single point.

Efficient algorithms for unconstrained global optimization use derivatives and partial derivatives to find local optima, to point out the direction in which moving from the current point does the most to increase or decrease the function. Such derivatives can sometimes be computed analytically, or they can be estimated numerically by taking the difference between values of nearby points. A variety of steepest descent and conjugate gradient methods to find local optima have been developed, similar in many ways to numerical root-finding algorithms. It is a good idea to try out several different methods on any given optimization problem. For this reason, we recommend experimenting with the implementations below before attempting to implement your own method. Clear descriptions of these algorithms are provided in several numerical algorithms books, in particular [PFTV86]. For constrained optimization, finding points that satisfy all the constraints is often the difficult problem. One approach is to use a method for unconstrained optimization, but add a penalty according to how many constraints are violated. Determining the right penalty function is problem-specific, but it often makes sense to vary the penalties as optimization proceeds. At the end, the penalties should be very high to ensure that all constraints are satisfied. Simulated annealing is a fairly robust and simple approach to constrained optimization, particularly when we are optimizing over combinatorial structures (permutations, graphs, subsets) instead of continuous functions. Techniques for simulated annealing are described in Section

.

Implementations: Several of the Collected Algorithms of the ACM are Fortran codes for unconstrained optimization, most notably Algorithm 566 [MGH81], Algorithm 702 [SF92], and Algorithm 734 [Buc94]. Algorithm 744 [Rab95] does unconstrained optimization in Lisp. They are available from Netlib (see Section ). Also check out the selection at GAMS, the NIST Guide to Available Mathematical Software, at http://gams.nist.gov. NEOS (Network-Enabled Optimization System) provides a unique service, the opportunity to solve your problem on computers and software at Argonne National Laboratory, over the WWW. Linear programming and unconstrained optimization are both supported. This is worth checking out at file:///E|/BOOK/BOOK3/NODE140.HTM (3 of 4) [19/1/2003 1:30:21]

Constrained and Unconstrained Optimization

http://www.mcs.anl.gov/home/otc/Server/ when you need a solution instead of a program. General purpose simulated annealing implementations are available and probably are the best place to start experimenting with this technique for constrained optimization. Particularly popular is Adaptive Simulated Annealing (ASA), written in C and retrievable via anonymous ftp from ftp.alumni.caltech.edu [131.215.139.234] in the /pub/ingber directory. To get on the ASA mailing list send e-mail to [email protected]. Genocop, by Zbigniew Michalewicz [Mic92], is a genetic algorithm-based program for constrained and unconstrained optimization, written in C. I tend to be quite skeptical of genetic algorithms (see Section ), but many people find them irresistible. Genocop is available from ftp://ftp.uncc.edu/coe/evol/ for noncommercial purposes. Notes: Steepest-descent methods for unconstrained optimization are discussed in most books on numerical methods, including [PFTV86, BT92]. Unconstrained optimization is the topic of several books, including [Bre73, Fle80]. Simulated annealing was devised by Kirkpatrick et. al. [KGV83] as a modern variation of the Metropolis Both use Monte Carlo techniques to compute the minimum energy state of a algorithm [MRRT53]. system. Good expositions on simulated annealing include [AK89]. Genetic algorithms were developed and popularized by Holland [Hol75, Hol92]. Expositions on genetic algorithms include [Gol89, Koz92, Mic92]. Tabu search [Glo89a, Glo89b, Glo90] is yet another heuristic search procedure with a devoted following. Related Problems: Linear programming (see page

), satisfiability (see page

).

Next: Linear Programming Up: Numerical Problems Previous: Determinants and Permanents Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE140.HTM (4 of 4) [19/1/2003 1:30:21]

Linear Programming

Next: Random Number Generation Up: Numerical Problems Previous: Constrained and Unconstrained Optimization

Linear Programming

Input description: A set S of n linear inequalities on m variables . linear optimization function

,

, and a

Problem description: Which variable assignment X' maximizes the objective function f while satisfying all inequalities S? Discussion: Linear programming is the most important problem in mathematical optimization and operations research. Applications include: ●

●

Resource allocation - We seek to invest a given amount of money so as to maximize our return. Our possible options, payoffs, and expenses can usually be expressed as a system of linear inequalities, such that we seek to maximize our possible profit given the various constraints. Very large linear programming problems are routinely solved by airlines and other corporations. Approximating the solution of inconsistent equations - A set of m linear equations on n variables , , is overdetermined if m > n. Such overdetermined systems are often inconsistent, meaning that no assignment of variables simultaneously solves all the equations. To find the

file:///E|/BOOK/BOOK3/NODE141.HTM (1 of 5) [19/1/2003 1:30:23]

Linear Programming

●

variable assignment that best fits the equations, we can replace each variable by and solve the new system as a linear program, minimizing the sum of the error terms. Graph algorithms - Many of the standard graph problems described in this book, such as shortest paths, bipartite matching, and network flow, can all be solved as special cases of linear programming. Most of the rest, including traveling salesman, set cover, and knapsack, can be solved using integer linear programming.

The standard algorithm for linear programming is called the simplex method. Each constraint in a linear programming problem acts like a knife that carves away a region from the space of possible solutions. We seek the point within the remaining region that maximizes (or minimizes) f(X). By appropriately rotating the solution space, the optimal point can always be made to be the highest point in the region. Since the region (simplex) formed by the intersection of a set of linear constraints is convex, we can find the highest point by starting from any vertex of the region and walking to a higher neighboring vertex. When there is no higher neighbor, we are at the highest point. While the basic simplex algorithm is not too difficult to program, there is a considerable art to producing an efficient implementation capable of solving large linear programs. For example, large programs tend to be sparse (meaning that most inequalities use few variables), so sophisticated data structures must be used. There are issues of numerical stability and robustness, as well as which neighbor we should walk to next (so called pivoting rules). Finally, there exist sophisticated interior-point methods, which cut through the interior of the simplex instead of walking along the outside, that beat simplex in many applications. The bottom line on linear programming is this: you are much better off using an existing LP code than writing your own. Further, you are much better off paying money than surfing the net. Linear programming is one algorithmic problem of such economic importance that commercial implementations are far superior to free versions. Issues that arise in linear programming include: ●

Do any variables have integrality constraints? - It is impossible to send 6.54 airplanes from New York to Washington each business day, even if that value maximizes profit according to your model. Such variables often have natural integrality constraints. A linear program is called an integer program when all its variables have integrality constraints, or a mixed integer progam if some of them do. Unfortunately, it is NP-complete to solve integer or mixed programs to optimality. However, there are techniques for integer programming that work reasonably well in practice. Cutting plane techniques solves the problem first as a linear program, and then adds extra constraints to enforce integrality around the optimal solution point before solving it again. After a sufficient number of iterations, the optimum point of the resulting linear program matches that of the original integer program. As with most exponential-time algorithms, run times for integer

file:///E|/BOOK/BOOK3/NODE141.HTM (2 of 5) [19/1/2003 1:30:23]

Linear Programming

●

programming depend upon the difficulty of the problem instance and are unpredictable. If they do not make progress quickly, they are unlikely make much progress over longer periods of time. Therefore, if you have multiple implementations available, it may well pay to try the same problem using different codes in the hopes that one can complete in a reasonable amount of time. What if my optimization function or constraints are not linear? - In least-squares curve fitting, we seek the line that best approximates a set of points by minimizing the sum of squares of the distance between each point and the line. In formulating this as a mathematical program, the natural objective function is no longer linear, but quadratic. Unfortunately, quadratic programming is NP-complete, even without integer variables. There are three possible courses of action when you must solve a nonlinear program. The best is to see if you can model it in some other way, as is the case with least-squares fitting. The second is to try to track down special codes for quadratic programming, which do exist. Finally, you can model your problem as a constrained or unconstrained optimization problem and try to solve

●

it with the codes discussed in Section . What if my model does not match the input format of my LP solver? - Many linear programming implementations accept models only in so-called standard form, where all variables are constrained to be nonnegative, the object function must be minimized, and all constraints must be equalities (instead of inequalities). Do not fear. There exist standard transformations to map arbitrary LP models into standard form. To convert a maximization problem to a minimization one, simply multiply each coefficient of the objective function by -1. The remaining problems can be solved by adding slack variables to the model. See any textbook on linear programming for details.

Implementations: A very useful resource on solving linear programs is the USENET frequently asked question (FAQ) list, maintained by John W. Gregory. In particular, it provides a list of available codes with descriptions of experiences. Check out the plaintext version at ftp://rtfm.mit.edu/pub/usenet/sci.answers/linear-programming-faq or a slicker WWW version at http://www.skypoint.com/ ashbury/linear-programming-faq.html. The noncommercial code of choice appears to be lp_solve, written in ANSI C by Michel Berkelaar, who has solved problems as large as 30,000 variables and 50,000 constraints. Lp_solve can also handle (smaller) integer and mixed-integer problems. It is available by anonymous ftp from ftp://ftp.es.ele.tue.nl/pub/lp_solve but is not in the public domain. A user community for lp_solve exists, which has ported it to a variety of different platforms. NEOS (Network-Enabled Optimization System) provides a unique service, an opportunity to solve your problem on computers and software at Argonne National Laboratory via the WWW. Linear programming and unconstrained optimization are both supported. This is worth checking out at http://www.mcs.anl.gov/home/otc/Server/ if you need an answer instead of a program. If you are serious about solving large linear programs, you likely need a commercial implementation. file:///E|/BOOK/BOOK3/NODE141.HTM (3 of 5) [19/1/2003 1:30:23]

Linear Programming

The book [MW93] provides an overview of commercial linear programming systems, online at http://www.mcs.anl.gov/home/otc/Guide/SoftwareGuide/index.html. Surveys of commercial LP codes appear in [SR95, Sha93] and in the linear programming FAQ. I have heard good things from various people about CPLEX and AMPL, but do your own research before spending money. For low-dimensional linear programming problems, computational geometry algorithms can outperform more general LP codes. See ftp://icemcfd.com/pub/linprog.a for a C language implementation of Seidel's randomized incremental LP algorithm, by Mike Hohmeyer. Algorithm 551 [Abd80] and Algorithm 552 [BR80] of the Collected Algorithms of the ACM are simplexbased codes for solving overdetermined systems of linear equations, in Fortran. details.

See Section

for

Pascal implementations of the revised and dual simplex methods for linear programming, as well as cutting plane and explicit enumeration algorithms for integer programming, are provided in [SDK83]. See Section

. These are likely to work only for small problems.

Sedgewick [Sed92] provides a bare bones implementation of the simplex algorithm in C++. See Section for details. Notes: Good expositions on the simplex and ellipsoid algorithms for linear programming include [PS82, Chv83]. Expositions on low-dimensional linear programming include [PS85]. For an implementationoriented exposition on linear and integer programming, with references to experimental work, see [SDK83]. The need for optimization via linear programming arose in logistics problems in World War II. The simplex algorithm was invented by George Danzig in 1947 [Dan63]. Klee and Minty [KM72] proved that the simplex algorithm is exponential in worst case, but it is very efficient in practice. Khachian's ellipsoid algorithm [Kha79] proved that linear programming was polynomial in 1979. Karmarkar's algorithm [Kar84] is an interior-point method that has proven to be both a theoretical and practical improvement of the ellipsoid algorithm, as well as a challenge for the simplex method. Linear programming is P-complete under log-space reductions [DLR79]. This makes it unlikely to have an NC parallel algorithm, where a problem is in NC iff it can be solved on a PRAM in polylogarithmic time using a polynomial number of processors. Any problem that is P-complete under log-space reduction cannot be in NC unless P=NC. See [GHR95] for a thorough exposition of the theory of Pcompleteness, including an extensive list of P-complete problems. Related Problems: Constrained and unconstrained optimization (see page file:///E|/BOOK/BOOK3/NODE141.HTM (4 of 5) [19/1/2003 1:30:23]

), network flow (see page

Linear Programming

).

Next: Random Number Generation Up: Numerical Problems Previous: Constrained and Unconstrained Optimization Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE141.HTM (5 of 5) [19/1/2003 1:30:23]

Random Number Generation

Next: Factoring and Primality Testing Up: Numerical Problems Previous: Linear Programming

Random Number Generation

Input description: Nothing, or perhaps a seed. Problem description: Generate a sequence of random integers. Discussion: Random number generation forms the foundation behind such standard algorithmic techniques as simulated annealing and Monte Carlo integration. Discrete event simulations, used to model everything from transportation systems to casino poker, all run on streams of random numbers. Initial passwords and cryptographic keys are typically generated randomly. New developments in randomized algorithms for graph and geometric problems are revolutionizing these fields and establishing randomization as one of the fundamental ideas of computer science. Unfortunately, generating random numbers is a task that looks a lot easier than it really is, primarily because it is fundamentally impossible to produce truly random numbers on any deterministic device. Von Neumann [vN63] said it best: ``Anyone who considers arithmetical methods of producing random digits is, of course, in a state of sin.'' All we can hope for are pseudorandom numbers, a stream of numbers that appear as if they were generated randomly. There can be serious consequences to using a bad random number generator. For example, the security of file:///E|/BOOK/BOOK4/NODE142.HTM (1 of 5) [19/1/2003 1:30:27]

Random Number Generation

an Internet password scheme was recently invalidated with the discovery that its keys were produced using a random number generator of such small period that brute-force search quickly exhausted all possible passwords. The accuracy of simulations is regularly compromised or invalidated by poor random number generation. Bottom line: This is an area where people shouldn't mess around, but they do. Issues to think about include: ●

●

●

Should my program use the same ``random'' numbers each time it runs? - A poker game that deals you the exact same hand each time you play quickly loses interest. One common solution is to use the lower-order bits of the machine clock as a seed or source for random numbers, so that each time the program runs it does something different. Such methods are perhaps adequate for games, but not for serious simulations. There are liable to be periodicities in the distribution of random numbers whenever calls are made in a loop. Also, debugging is seriously complicated by the fact that the results are not repeatable. If the program crashes, you cannot go back and discover why. One possible compromise is to use a deterministic pseudorandom number generator, but write the current seed to a file between runs. During debugging, this file can be overwritten with a fixed initial value or seed. How good is my compiler's built-in random number generator? - If you need uniformly generated random numbers, and you are not going to bet the farm on the accuracy of your simulation, my recommendation is simply to use what your compiler provides. Your best opportunity to mess it up is with a bad choice of starting seed, so read the manual for its recommendations. If you are going to bet the farm on the quality of your simulation, you had better test your random number generator. Be aware that it is very difficult to eyeball the results and decide whether the output is really random. This is because people have very skewed ideas of how random sources should behave and often see patterns that don't really exist. To evaluate a random number generator, several different tests should be used and the statistical significance of the results established. Such tests are implemented in plab and DIEHARD (discussed below) and explained in [Knu81]. What if I have to implement my own random number generator? - The algorithm of choice is the linear congruential generator. It is fast, simple, and (if instantiated with the right constants) gives reasonable pseudorandom numbers. The nth random number is a function of the (n-1)st random number:

In theory, linear congruential generators work the same way roulette wheels do. The long path of the ball around and around the wheel (captured by ) ends in one of a relatively small number of bins, the choice of which is extremely sensitive to the length of the path (captured by the truncation of the ). A substantial theory has been developed to select the constants a, c, m, and file:///E|/BOOK/BOOK4/NODE142.HTM (2 of 5) [19/1/2003 1:30:27]

. The period length

Random Number Generation

●

is largely a function of the modulus m, which is typically constrained by the word length of the , a = 1366, c=150889, machine. A presumably safe choice for a 32-bit machine would be and m=714025. Don't get creative and change any of the constants, unless you use the theory or run tests to determine the quality of the resulting sequence. What if I don't want large, uniformly distributed random integers? - The linear congruential generator produces a uniformly distributed sequence of large integers, which can be scaled to produce other uniform distributions. For uniformly distributed real numbers between 0 and 1, use . Note that 1 cannot be realized this way, although 0 can. If you want uniformly distributed integers between l and h, use . Generating random numbers according to a given nonuniform distribution can be a tricky business. The most reliable way to do this correctly is the acceptance-rejection method. Suppose we bound the desired probability distribution function or geometric region to sample from a box and then select a random point p from the box. This point can be selected by p by generating the x and y coordinates independently, at random. If this p is within the distribution, or region, we can return p as selected at random. If p is in the portion of the box outside the region of interest, we throw it away and repeat with another random point. Essentially, we throw darts at random and report those that hit the target. This method is correct, but it can be slow. If the volume of the region of interest is small relative to that of the box, most of our darts will miss the target. Efficient generators for Gaussian and other special distributions are described in the references and implementations below.

●

Be cautious about inventing your own technique, however, since it can be tricky to obtain the right probability distribution. For example, an incorrect way to select points uniformly from a circle of radius r would be to generate polar coordinates and select an angle from 0 to and a displacement between 0 and r, both uniformly at random. In such a scheme, half the generated points will lie within r/2 of the radius, when only one-fourth of them should be! This is a substantial enough difference to seriously skew the results, while being subtle enough that it might escape detection. How long should I run my Monte Carlo simulation to get the best results? - It makes sense that the longer you run a simulation, the more accurately the results will approximate the limiting distribution, thus increasing accuracy. However, this is only true until you exceed the period, or cycle length, of your random number generator. At that point, your sequence of random numbers repeats itself, and further runs generate no additional information. Check the period length of your generator before you jack up the length of your simulation. You are liable to be very surprised by what you learn.

Implementations: An excellent WWW page on random number generation and stochastic simulation is available at http://random.mat.sbg.ac.at/others/. It includes pointers to papers and literally dozens of implementations of random number generators. From there are accessible pLab [Lee94] and DIEHARD, systems for testing the quality of random number generators. file:///E|/BOOK/BOOK4/NODE142.HTM (3 of 5) [19/1/2003 1:30:27]

Random Number Generation

The Stanford Graphbase (see Section based on the recurrence . period of at least

) contains a machine-independent random number generator . With the proper initialization, this generator has a

Algorithm 488 [Bre74], Algorithm 599 [AKD83], and Algorithm 712 [Lev92] of the Collected Algorithms of the ACM are Fortran codes for generating random numbers according to several probability distributions, including normal, exponential, and Poisson distributions. They are available from Netlib (see Section

).

Sim++ is a library of routines for implementing discrete event simulations, built by Robert Cubert and Paul Fishwick, of the University of Florida. It contains random number generators for a variety of different distributions, including uniform, exponential, and normal. Check out http://www.cis.ufl.edu/ fishwick/simpack/simpack.html if you need a random number generator to control a simulation. Fishwick's book [Fis95] describes model design using SimPack. ) provides a comprehensive random source in C++ for generating random bits, LEDA (see Section integers, and double precision reals. Sedgewick [Sed92] provides simple implementations of linear and additive congruential generators in C++. See Section

for details.

) is an algorithm animation system for UNIX and X-windows, which includes XTango (see Section an animation illustrating the uniformity of random number generation. Notes: Knuth [Knu81] has a thorough and interesting discussion of random number generation, which I heartily recommend. He presents the theory behind several methods, including the middle square and shift-register methods we have not described here, as well as a detailed discussion of statistical tests for validating random number generators. Another good source is [PFTV86] - our recommended constants for the linear congruential generator are drawn from here. Comparisons of different random number generators in practice include [PM88]. Tables of random numbers appear in most mathematical handbooks, as relics from the days before there was ready access to computers. Most notable is [RC55], which provides one million random digits. The deep relationship between randomness, information, and compressibility is explored within the theory of Kolmogorov complexity, which measures the complexity of a string by its compressibility. Truly random strings are incompressible. The string of seemingly random digits of cannot be random under this definition, since the entire sequence is defined by any program implementing a series expansion for . Li and Vitáni [LV93] provide a thorough introduction to the theory of Kolmogorov complexity. file:///E|/BOOK/BOOK4/NODE142.HTM (4 of 5) [19/1/2003 1:30:27]

Random Number Generation

Related Problems: Constrained and unconstrained optimization (see page (see page

), generating subsets (see page

), generating permutations

), generating partitions (see page

).

Next: Factoring and Primality Testing Up: Numerical Problems Previous: Linear Programming Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE142.HTM (5 of 5) [19/1/2003 1:30:27]

Factoring and Primality Testing

Next: Arbitrary-Precision Arithmetic Up: Numerical Problems Previous: Random Number Generation

Factoring and Primality Testing

Input description: An integer n. Problem description: Is n a prime number, and if not what are the factors of n? Discussion: The dual problems of factoring integers and testing primality have surprisingly many applications for a problem long suspected of being only of mathematical interest. The security of the RSA public-key cryptography system (see Section ) is based on the computational intractability of factoring large integers. As a more modest application, hash table performance typically improves when the table size is a prime number. To get this benefit, an initialization routine must identify a prime near the desired table size. Finally, prime numbers are just interesting to play with. It is no coincidence that programs to generate large primes often reside in the games directory of UNIX systems. Although factoring and primality testing are related problems, algorithmically they are quite different. There exist algorithms that can demonstrate that an integer is composite (i.e. not prime) without actually giving the factors. To convince yourself of the plausibility of this, note that you can demonstrate the compositeness of any nontrivial integer whose last digit is 0, 2, 4, 5, 6, or 8 without doing the actual division.

file:///E|/BOOK/BOOK4/NODE143.HTM (1 of 4) [19/1/2003 1:30:28]

Factoring and Primality Testing

The simplest algorithm for both of these problems is brute-force trial division. To factor n, compute the remainder of n/i for all . The prime factorization of n will contain at least one instance of , unless n is prime. Make sure you handle the multiplicities correctly and every i such that . account for any primes larger than Such algorithms can be sped up by using a precomputed table of small primes to avoid testing all possible i. Surprisingly large numbers of primes can be represented in surprisingly little space by using ). A bit vector of all odd numbers less than 1,000,000 fits in under 64 bit vectors (see Section kilobytes. Even tighter encodings become possible by eliminating all multiples of three and other small primes. Considerably faster factoring algorithms exist, whose correctness depends upon more substantial number theory. The fastest known algorithm, the number field sieve, uses randomness to construct a system of congruences, the solution of which usually gives a factor of the integer. Integers with as many at 128 digits have been factored using this method, although such feats require enormous amounts of computation. Randomized algorithms make it much easier to test whether an integer is prime. Fermat's little theorem states that for all a, when n is prime. Suppose we pick a random value of and . If this residue is not 1, we have just proven that n cannot be prime. compute the residue of Such randomized primality tests are very efficient. PGP (see Section hundreds of these tests in minutes, for use as cryptographic keys.

) finds 300+ digit primes using

Although the primes are scattered in a seemingly random way throughout the integers, there is some regularity to their distribution. The prime number theorem states that the number of primes less than n, commonly denoted by , is approximately . Further, there are never large gaps between primes, so in general, one would expect to examine about integers if one wanted to find the first prime larger than n. This distribution and the fast randomized primality test explain how PGP can find such large primes so quickly. Implementations: My first choice for factoring or primality testing applications would be PARI, a system capable of handling complex number-theoretic problems on integers with up to 300,000 decimal digits, as well as reals, rationals, complex numbers, polynomials, and matrices. It is written mainly in C, with assembly code for inner loops on major architectures, and includes more than 200 special predefined mathematical functions. PARI can be used as a library, but it also possesses a calculator mode that gives instant access to all the types and functions. The main advantage of PARI is its speed. On a Unix platform, it is between 5 to 100 times faster than Maple or Mathematica, depending on the applications. PARI is available for PC, Amiga, Macintosh, and most Unix platforms by anonymous ftp at ftp://megrez.ceremab.u-bordeaux.fr/pub/pari/.

file:///E|/BOOK/BOOK4/NODE143.HTM (2 of 4) [19/1/2003 1:30:28]

Factoring and Primality Testing

A Mathematica implementation by Ilan Vardi of Lenstra's elliptic curve method of factorization is available in Packages/NumberTheory/FactorIntegerECM.m of the standard Mathematica distribution and MathSource. It is designed to find prime factors of up to about 18 digits in reasonable time, extending Mathematica's ability to factor all numbers of up to 40 digits. It is faster when factoring the product of small primes. Notes: Bach and Shallit's book [BS96] is the most comprehensive reference on computational number theory, while Adleman's excellent survey [Adl94a] describes the state of the art, as well as open problems. Good expositions on modern algorithms for factoring and primality testing include [CLR90]. The Miller-Rabin [Mil76, Rab80] randomized primality testing algorithm eliminates problems with Carmichael numbers, which are composite integers that always satisfy Fermat's theorem. The best algorithms for integer factorization include the quadratic-sieve [Pom84] and the elliptic-curve methods [HWL87]. Mechanical sieving devices provided the fastest way to factor integers surprisingly far into the computing era. See [SWM95] for a fascinating account of one such device, built during World War I. Hand-cranked, it proved the primality of in fifteen minutes of sieving time. An important problem in computational complexity theory is whether P = NP co-NP. The decision problem ``is n a composite number?'' is perhaps the best candidate for a counterexample. By exhibiting the factors of n, it is trivially in NP. It can be shown to be in co-NP, since every prime has a short proof of its primality [Pra75]. However, there is no evidence it is in P. For more information on complexity classes, see [GJ79, Joh90]. A group headed by Arjen Lenstra has regularly broken records for general-purpose integer factoring, using an Internet-distributed implementation of the quadratic sieve factoring method. The June 1993 factorization of RSA-120 took approximately 830 MIP-years of computation. The April 1994 factorization of RSA-129, famous for appearing in the original RSA paper [RSA78], was factored in eight months using over 1,600 computers. This was particularly noteworthy because in [RSA78] they had originally predicted such a factorization would take 40 quadrillion years using 1970s technology. Related Problems: Cryptography (see page

), high precision arithmetic (see page

).

Next: Arbitrary-Precision Arithmetic Up: Numerical Problems Previous: Random Number Generation

file:///E|/BOOK/BOOK4/NODE143.HTM (3 of 4) [19/1/2003 1:30:28]

Factoring and Primality Testing

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE143.HTM (4 of 4) [19/1/2003 1:30:28]

Arbitrary-Precision Arithmetic

Next: Knapsack Problem Up: Numerical Problems Previous: Factoring and Primality Testing

Arbitrary-Precision Arithmetic

Input description: Two very large integers, x and y. Problem description: What is x+y, x-y,

, and x / y?

Discussion: Any programming language whose level rises above basic assembler supports single- and perhaps double-precision integer/real addition, subtraction, multiplication, and division. But what if we wanted to represent the national debt of the United States in pennies? One trillion dollars worth of pennies requires 15 decimal digits, which is far more than can fit into a 32-bit integer. In other applications much larger integers are needed. The RSA algorithm for public-key cryptography requires integer keys of at least 100 digits to achieve any level of security, and 1000 digits are recommended. Experimenting with number-theoretic conjectures for fun or research always requires playing with large numbers. I once solved a minor open problem [GKP89] by performing an exact computation on the integer

.

What should you do when you need large integers? ●

Am I solving a problem instance requiring large integers, or do I have an embedded application? -

file:///E|/BOOK/BOOK4/NODE144.HTM (1 of 5) [19/1/2003 1:30:30]

Arbitrary-Precision Arithmetic

If you just need the answer to a specific problem with large integers, such as in the number theory application above, you would be well advised to consider using a computer algebra system like Maple or Mathematica. These use arbitrary-precision arithmetic as a default and use nice Lisplike programming languages as a front end, together often reducing your problem to a 5 to 10 line program.

●

●

If instead you have an embedded application requiring high-precision arithmetic, you would be well advised to use an existing library. In addition to the four basic operations, you are likely to get additional functions for computing things like greatest common divisor in the bargain. See the implementations below for details. Do I need high- or arbitrary-precision arithmetic? - Is there an upper bound on how big your integers can get, or do you really need arbitrary-precision, i.e. unbounded. This determines whether you can use a fixed-length array to represent your integers as opposed to a linked-list of digits. The array is likely to be simpler and will not be a constraint in most applications. What base should I do arithmetic in? - It is conceptually simplest to implement your own highprecision arithmetic package in decimal and represent each integer as a string of base-10 digits, at one digit per node. However, it is far more efficient to use a higher base, ideally the square root of the largest integer supported fully by hardware arithmetic. Why? The higher the base, the fewer digits we need to represent the number (compare 64 decimal with 1000000 binary). Since hardware addition usually takes one clock cycle independent of the actual numbers, best performance is achieved using the highest base. The reason for limiting us to is that in performing high-precision multiplication, we will multiply two of these ``digits'' together and need to avoid overflow.

●

The only complication of using a larger base is that integers must be converted to and from base10 for input and output, but the conversion is easily performed once all four high-precision arithmetical operations are supported. How low-level are you willing to get for fast computation? Hardware addition is much faster than a subroutine call, so you are going to take a significant hit on speed whenever your package is used where low-precision arithmetic suffices. High-precision arithmetic is one of few problems in this book where inner loops in assembly language can be the right idea to speed things up. Finally, using bit-level masking and shift operations instead of arithmetical operations can be a win.

The algorithm of choice for each of the five basic arithmetic operations is as follows: ●

●

Addition - The basic schoolhouse method of lining up the decimal points and then adding the digits from right to left with carries works in time linear in the number of digits. More sophisticated carry-look-ahead parallel algorithms are available for low-level hardware implementation. Hopefully they are used on your chip for low-precision addition. Subtraction - Depending upon the sign bits of the numbers, subtraction can be a special case of

file:///E|/BOOK/BOOK4/NODE144.HTM (2 of 5) [19/1/2003 1:30:30]

Arbitrary-Precision Arithmetic

●

●

●

addition: (A - (-B)) = (A+B). The tricky part of subtraction is performing the borrow. This can be simplified by always subtracting from the number with the larger absolute value and adjusting the signs afterwards, so we can be certain there will always be something to borrow from. Multiplication - The simplest method of repeated addition will take exponential time on large integers, so stay away. The digit-by-digit schoolhouse method is reasonable to program and will work much better, presumably well enough for your application. On very large integers, Karatsuba's divide-and-conquer algorithm (cited in the notes) wins. Dan Grayson, author of Mathematica's arbitrary-precision arithmetic, found that the switch-over happened at well under 100 digits. Even faster on very large integers is an algorithm based on Fourier transforms. A . discussion of such algorithms appears in Section Division - Repeated subtraction will take exponential time, so the easiest reasonable algorithm to use is the long-division method you hated in school. This is a far more complicated algorithm than needed for the other operations, requiring arbitrary-precision multiplication and subtraction as subroutines, as well as trial and error to determine the correct digit to use at each position of the quotient. In fact, integer division can be reduced to integer multiplication, although in a nontrivial way, so if you are implementing asymptotically fast multiplication, you can reuse that effort in long division. See the references below for details. Exponentiation - We can compute in the obvious manner using b-1 multiplications, but a better . By repeatedly squaring the results of our partial way is to exploit the fact that product, we can escape using for a discussion of this algorithm.

multiplications, a big win when b is large.

See Section

High- but not arbitrary-precision arithmetic can be conveniently performed using the Chinese remainder theorem and modular arithmetic. The Chinese remainder theorem states that an integer between 1 and is uniquely determined by its set of residues mod , where each are relatively prime integers. Addition, subtraction, and multiplication (but not division) can be supported using such residue systems, with the advantage that large integers can be manipulated without complicated data structures. Many of these algorithms for computations on long integers can be directly applied to computations on polynomials. See the references for more details. A particularly useful algorithm is Horner's rule for fast polynomial evaluation. When is blindly evaluated term by term, multiplications will be performed. Much better is observing that , the evaluation of which uses only a linear number of operations. Implementations: All major commercial computer algebra systems incorporate high-precision arithmetic, including Maple, Mathematica, Axiom, and Macsyma. If you have access to one of these, this is your best option for a quick, nonembedded application. The rest of this section focuses on source code available for embedded applications.

file:///E|/BOOK/BOOK4/NODE144.HTM (3 of 5) [19/1/2003 1:30:30]

Arbitrary-Precision Arithmetic

PARI, developed by Henri Cohen and his colleagues in France, is a system capable of handling complex number-theoretic problems on integers with up to 300,000 decimal digits, as well as reals, rationals, complex numbers, polynomials, and matrices. It is probably the most powerful free software available for number theory. Written mainly in C (with assembly-language code for speed-critical routines), it includes more than 200 special predefined mathematical functions. PARI can be used as a library, but it possesses also a powerful calculator mode that gives instant access to all the types and functions. The main advantage of PARI is its speed. On a Unix platform, it runs between 5 to 100 times faster than Maple or Mathematica, depending on the applications. PARI is available for PC, Amiga, Macintosh, and most Unix platforms by anonymous ftp at ftp://megrez.ceremab.u-bordeaux.fr/pub/pari/. Algorithm 693 [Smi91] of the Collected Algorithms of the ACM is a Fortran implementation of floatingpoint, multiple-precision arithmetic. See Section

.

An implementation of arbitrary-precision integer and rational arithmetic in C++ is embedded in LEDA (see Section ), including GCD, square roots, and logarithms as well as the basic four operations. Sparc assembler code is used for certain time-critical functions. Implementations in C of a high-precision calculator with all four elementary operations appear in [BR95]. The authors use base 10 for arithmetic and arrays of digits to represent long integers, with short integers as array indices, thus limiting computations to 32,768 digits. The code for these algorithms is printed in the text and available on disk for a modest fee. Bare bones implementations in C of high-precision multiplication and in Pascal of such special functions as logarithm and arctangent appear in [GBY91]. See Section

for further details. Sedgewick [Sed92]

provides a bare bones implementation of polynomial arithmetic in C++. See Section

for details.

Notes: Knuth [Knu81] is the primary reference on algorithms for all basic arithmetic operations, including implementations of them in the MIX assembly language. Bach and Shallit [BS96] provide a more recent treatment of computational number theory. Expositions on the -time divide-and-conquer algorithm for multiplication [KO63] include [AHU74, Man89]. An FFT-based algorithm multiplies two n-bit numbers in time and is due to Schönhage and Strassen [SS71]. Expositions include [AHU74, Knu81]. The reduction between integer division and multiplication is presented in [AHU74, Knu81]. Good expositions of algorithms for modular arithmetic and the Chinese remainder theorem include [AHU74, CLR90]. A good exposition of circuit-level algorithms for elementary arithmetic algorithms is [CLR90].

file:///E|/BOOK/BOOK4/NODE144.HTM (4 of 5) [19/1/2003 1:30:30]

Arbitrary-Precision Arithmetic

Euclid's algorithm for computing the greatest common divisor of two numbers is perhaps the oldest interesting algorithm. Expositions include [CLR90, Man89]. Related Problems: Factoring integers (see page

), cryptography (see page

).

Next: Knapsack Problem Up: Numerical Problems Previous: Factoring and Primality Testing Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE144.HTM (5 of 5) [19/1/2003 1:30:30]

Knapsack Problem

Next: Discrete Fourier Transform Up: Numerical Problems Previous: Arbitrary-Precision Arithmetic

Knapsack Problem

Input description: A set of items capacity C. Problem description: Find the subset i.e. all the items fit in a knapsack of size C.

, where item i has size

that maximizes the value of

and value

. A knapsack

given that

;

Discussion: The knapsack problem arises whenever there is resource allocation with financial constraints. Given a fixed budget, how do you select what things you should buy. Everything has a cost and value, so we seek the most value for a given cost. The term knapsack problem invokes the image of the backbacker who is constrained by a fixed-size knapsack and so must fill it only with the most useful items. The typical formulation in practice is the 0/1 knapsack problem, where each item must be put entirely in the knapsack or not included at all. Objects cannot be broken up arbitrarily, so its not fair taking one can of coke from a six-pack or opening the can to take just a sip. It is this 0/1 property that makes the knapsack problem hard, for a simple greedy algorithm finds the optimal selection whenever we are allowed to subdivide objects arbitrarily. For each item, we could compute its ``price per pound'', and take as much of the most expensive item until we have it all or the knapsack is full. Repeat with the next file:///E|/BOOK/BOOK4/NODE145.HTM (1 of 4) [19/1/2003 1:30:33]

Knapsack Problem

most expensive item, until the knapsack is full. Unfortunately, this 0/1 constraint is usually inherent in most applications. Issues that arise in selecting the best algorithm include: ●

●

Does every item have the same cost/value or the same size? - If each item is worth the same amount to us as any other item, say $1, I maximize my value by taking the greatest number of items. In this case the optimal solution is to sort the items in order of increasing size and insert them into the knapsack in this order until nothing fits. The problem is solved similarly when each object has the same size but the costs are different. These are the easy cases of knapsack. Does each item have the same ``price per pound''? - In this case, our problem is equivalent to ignoring the price and just trying to minimize the amount of empty space left in the knapsack. Unfortunately, even this restricted version of the problem is NP-complete, and so we cannot expect an efficient algorithm that always solves the problem. Don't lose hope, however, because knapsack proves to be an ``easy'' hard problem, one that can usually be handled with the algorithms described below. An important special case of constant ``price-per-pound'' knapsack is the integer partition problem, presented in cartoon form in Figure

.

Figure: Integer partition is a variant of the Knapsack problem Here, we seek to partition the elements of S into two sets A and B such that , or alternately make the difference as small as possible. Integer partition can be thought of as bin packing with two equal-sized bins or knapsack with a capacity of half the total weight, so all three problems are closely related and NP-complete.

●

The constant `price-per-pound' knapsack problem is often called the subset sum problem, because given a set of numbers, we seek a subset that adds up to a specific target number, i.e. the capacity of our knapsack. Are all the sizes relatively small integers? - When the sizes of the items and the knapsack capacity C are all integers, there is an efficient dynamic programming algorithm that finds the optimal solution in time O(n C) and O(C) space. Whether this is good for you depends upon how big C is. For 1,000, this might be great, but not for 10,000,000. The algorithm works as follows: Let S' be a set of items, and let C[i,S'] be true if and only if there

file:///E|/BOOK/BOOK4/NODE145.HTM (2 of 4) [19/1/2003 1:30:33]

Knapsack Problem

is a subset of S' whose size adds up exactly to i. For the empty set, is false for . One by one we add a new item to S' and update the affected values of C[i,S']. Observe that iff either C[i,S'] or is true, since either we use in our subset or we , and updating don't. By performing n sweeps through all C elements, one for each , the array, we identify which sums of sizes can be realized. The knapsack solution is the largest realizable size. In order to reconstruct the winning subset, we must store the name of the item , and then scan backwards through number that turned C[i] from false to true, for each the array.

●

The dynamic programming formulation described above ignored the values of the items. To generalize the algorithm, add a field to each element of the array to store the value of the best subset to date summing up to i. We now update not only when C[i] turns from false to true, but when the sum of the cost of plus the cost of is better than the previous cost of C[i]. What if I have multiple knapsacks? - When there are multiple knapsacks, your problem is probably better thought of as a bin packing problem. See Section

.

When the knapsack capacity gets too large for dynamic programming, exact solutions can be found using integer programming or backtracking. A 0/1 integer variable is used to denote whether item i is present in the optimal subset. We maximize given the constraint that . Algorithms and implementations of integer and linear programming are discussed in Section

.

When exact solutions prove too costly to compute, heuristics should be used. The simple greedy heuristic inserts items according to the maximum `price per pound' rule, described above. Often this heuristic solution is close to optimal, but it can be arbitrarily bad depending upon the problem instance. The ``price per pound'' rule can also be used to reduce the size of the problem instance in exhaustive searchbased algorithms by eliminating ``cheap but heavy'' objects from future consideration. Another heuristic is based on scaling. Dynamic programming works well if the capacity of the knapsack is a reasonably small integer, say . But what if we have a problem with capacity ? We scale , round the size down to an integer, and then use dynamic down the sizes of all items by a factor of programming on the scaled items. Scaling works well in practice, especially when the range of sizes of items is not too large. Implementations: Martello and Toth's book [MT90a] comes with a disk of Fortran implementations of a variety of knapsack algorithms. This is likely the best source of code currently available. Algorithm 632 [MT85] of the Collected Algorithms of the ACM is a Fortran code for the 0/1 knapsack problem, with the twist that it supports multiple knapsacks. See Section

file:///E|/BOOK/BOOK4/NODE145.HTM (3 of 4) [19/1/2003 1:30:33]

.

Knapsack Problem

Pascal implementations of several knapsack algorithms, including backtracking and a refined greedy algorithm, are provided in [SDK83]. See Section

for details.

Notes: Martello and Toth's book [MT90a] and survey article [MT87] are the standard references on the knapsack problem, including most theoretical and experimental results. An excellent exposition on integer programming approaches to knapsack problems appears in [SDK83]. See [FP75a] for a computational study of algorithms for 0-1 knapsack problems. A polynomial-time approximation scheme is an algorithm that approximates the optimal solution of a problem in time polynomial in both its size and the approximation factor . This very strong condition implies a smooth tradeoff between running time and approximation quality. Good expositions on the polynomial-time approximation scheme [IK75] for knapsack and subset sum includes [Baa88, CLR90, GJ79, Man89]. The first algorithm for generalized public key encryption by Merkle and Hellman [MH78] was based on the hardness of the knapsack problem. See [Sch94] for an exposition. Related Problems: Bin packing (see page

), integer programming (see page

).

Next: Discrete Fourier Transform Up: Numerical Problems Previous: Arbitrary-Precision Arithmetic Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE145.HTM (4 of 4) [19/1/2003 1:30:33]

Discrete Fourier Transform

Next: Combinatorial Problems Up: Numerical Problems Previous: Knapsack Problem

Discrete Fourier Transform

Input description: A sequence of n real or complex values intervals from a function h. Problem description: The discrete Fourier transform H of h,

,

, sampled at uniform

, for

.

Discussion: Although computer scientists tend to be relatively unfamiliar with Fourier transforms, electrical engineers and signal processors eat them for breakfast. Functionally, Fourier transforms provide a way to convert samples of a standard time-series into the ``frequency domain''. This provides a ``dual'' representation of the function, in which certain operations become easier than in the time domain. Applications of Fourier transforms include: ●

●

Filtering - Taking the Fourier transform of a function is equivalent to representing it as the sum of sine functions. By eliminating undesirable high- and/or low-frequency components (i.e. dropping some of the sine functions) and taking an inverse Fourier transform to get us back into the time domain, we can filter an image to remove noise and other artifacts. For example, the sharp spike in the figure above describes the period of a single sine function that closely models the input data. Image Compression - A smoothed, filtered image contains less information than a noisy image,

file:///E|/BOOK/BOOK4/NODE146.HTM (1 of 4) [19/1/2003 1:30:34]

Discrete Fourier Transform

●

●

while retaining a similar appearance. Thus encoding the smoothed image will require fewer bits to represent than the original image. By eliminating the coefficients of sine functions that contribute relatively little to the image, we can further reduce the size of the image, at little cost in image fidelity. Convolution and Deconvolution - Fourier transforms can be used to efficiently compute convolutions of two sequences. A convolution is the pairwise product of elements from two different sequences, such as in multiplying two n-variable polynomials f and g or multiplying two long integers. Implementing the product directly takes , while suffices using the fast Fourier transform. Another example comes from image processing. Because a scanner measures the darkness of an image patch instead of a single point, the scanned input is always blurred. A reconstruction of the original signal can be obtained by deconvoluting the input signal with a Gaussian point-spread function. Computing the correlation of functions - The correlation function of two functions f(t) and g(t) is defined by

and can be easily computed using Fourier transforms. Note that if the two functions are similar in shape but one is shifted relative to the other (such as and ), the value of will be large at this shift offset . As an application, suppose that we want to detect whether there are any funny periodicities in our random number generator. We can generate a large series of random numbers, turn it into a time series (the ith number at time i), and take the Fourier transform of this series. Any funny spikes will correspond to potential periodicities. The discrete Fourier transform takes as input n complex numbers

, , corresponding to equally spaced points in a time series, and outputs n complex numbers , , each describing a sine function of given frequency. The discrete Fourier transform is defined by

and the inverse Fourier transform is defined by

which enables us move easily between h and H. Since the output of the discrete Fourier transform consists of n numbers, each of which can be computed using a formula on n numbers, they can be computed in time. The fast Fourier transform (FFT) is

file:///E|/BOOK/BOOK4/NODE146.HTM (2 of 4) [19/1/2003 1:30:34]

Discrete Fourier Transform

an algorithm that computes the discrete Fourier transform in . This is arguably the most important algorithm known, as measured by practical impact, for it opened the door to modern image processing. There are several different algorithms that call themselves FFTs, all of which are based on a divide-and-conquer approach. Essentially, the problem of computing the discrete Fourier transform on n points is reduced to computing two transforms on n/2 points each and is then applied recursively The FFT usually assumes that n is a power of two. If this is not the case for your data, you are usually better off padding your data with zeros to create elements rather than hunting for a more general code. Since many image processing systems have strong real-time constraints, FFTs are often implemented in hardware, or at least in assembly language tuned to the particular machine. Be aware of this possibility if the codes below prove too slow. Implementations: FFTPACK is a package of Fortran subprograms for the fast Fourier transform of periodic and other symmetric sequences, written by P. Swartzrauber. It includes complex, real, sine, cosine, and quarter-wave transforms. A C language translation of the main routines is also provided. FFTPACK resides on Netlib (see Section

) at http://www.netlib.org/fftpack.

Algorithm 545 [Fra79] of the Collected Algorithms of the ACM is an implementation of the fast Fourier transform optimizing virtual memory performance and written in Fortran. information.

See Section

for further

XTango (see Section ) is an algorithm animation system for UNIX and X-windows, which includes an interesting animation of the fast Fourier transform. A Pascal implementation of the fast Fourier transform for details, see Section

points appears in [MS91]. For more

. Sedgewick [Sed92] provides a bare bones implementation of the fast Fourier

transform in C++. See Section

for details.

Notes: Brigham [Bri74] is an excellent introduction to Fourier transforms and the FFT and is strongly recommended, as is the exposition in [PFTV86]. Expositions in algorithms texts on the fast Fourier transform include [AHU74, Baa88, CLR90, Man89]. Credit for inventing the fast Fourier transform is usually given to Cooley and Tukey [CT65], although it is not completely deserved. See [Bri74] for a complete history. An interesting divide-and-conquer algorithm for polynomial multiplication [KO63] does the job in

file:///E|/BOOK/BOOK4/NODE146.HTM (3 of 4) [19/1/2003 1:30:34]

Discrete Fourier Transform

time and is discussed in [AHU74, Man89]. An FFT-based algorithm that multiplies two n-bit time is due to Schönhage and Strassen [SS71] and is presented in [AHU74]. numbers in In recent years, wavelets have been proposed to replace Fourier transforms in filtering. See [Dau92] for an introduction to wavelets. Related Problems: Data compression (see page

), high-precision arithmetic (see page

Next: Combinatorial Problems Up: Numerical Problems Previous: Knapsack Problem Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE146.HTM (4 of 4) [19/1/2003 1:30:34]

).

Combinatorial Problems

Next: Sorting Up: A Catalog of Algorithmic Previous: Discrete Fourier Transform

Combinatorial Problems In this section, we consider several classic algorithmic problems of a purely combinatorial nature. These include sorting and permutation generation, both of which were among the first nonnumerical problems arising on electronic computers. Sorting, searching, and selection can all be classified in terms of operations on a partial order of keys. Sorting can be viewed as identifying or imposing the total order on the keys, while searching and selection involve identifying specific keys based on their position in the total order. The rest of this section deals with other combinatorial objects, such as permutations, partitions, subsets, calendars, and schedules. We are particularly interested in algorithms that rank and unrank combinatorial objects, i.e. that map each distinct object to and from a unique integer. Once we have rank and unrank operations, many other tasks become simple, such as generating random objects (pick a random number and unrank) or listing all objects in order (iterate from 1 to n and unrank). We conclude with the problem of generating graphs. Graph algorithms are more fully presented in subsequent sections. Books on general combinatorial algorithms, in this restricted sense, include: ●

●

●

●

●

Nijenhuis and Wilf [NW78] - This book specializes in algorithms for constructing basic combinatorial objects such as permutations, subsets, and partitions. Such algorithms are often very short but hard to locate and usually are surprisingly subtle. Fortran programs for all of the algorithms are provided, as well as a discussion of the theory behind each of them. See Section for details. Ruskey [Rus97] - On its completion, this manuscript in preparation will become the standard reference on generating combinatorial objects. A preview is available via the WWW at http://www-csc.uvic.ca/home/fruskey/cgi-bin/html/main.html . Knuth [Knu73a, Knu73b] - The standard reference on searching and sorting, with significant material on combinatorial objects such as permutations. Reingold, Nievergelt, Deo [RND77] - A comprehensive algorithms text with a particularly thorough treatment of combinatorial generation and search. Stanton and White [SW86] - An undergraduate combinatorics text with algorithms for generating permutations, subsets, and set partitions. It contains relevant programs in Pascal.

file:///E|/BOOK/BOOK4/NODE147.HTM (1 of 2) [19/1/2003 1:30:35]

Combinatorial Problems ●

Skiena [Ski90] - This description of Combinatorica, a library 230 Mathematica functions for generating combinatorial objects and graph theory (see Section ) provides a distinctive view of how different algorithms can fit together. Its author is uniquely qualified to write a manual on algorithm design.

● ● ● ● ● ● ● ● ● ●

Sorting Searching Median and Selection Generating Permutations Generating Subsets Generating Partitions Generating Graphs Calendrical Calculations Job Scheduling Satisfiability

Next: Sorting Up: A Catalog of Algorithmic Previous: Discrete Fourier Transform Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE147.HTM (2 of 2) [19/1/2003 1:30:35]

Sorting

Next: Searching Up: Combinatorial Problems Previous: Combinatorial Problems

Sorting

Input description: A set of n items. Problem description: Arrange the items in increasing order. Discussion: Sorting is the fundamental algorithmic problem in computer science. Learning the different sorting algorithms is like learning scales for a musician. Sorting is the first step in solving a host of other algorithm problems, as discussed in Section rules of algorithm design.

. Indeed, ``when in doubt, sort'' is one of the first

Sorting is also used to illustrate the standard paradigms of algorithm design. The result is that most programmers are familiar with many different sorting algorithms, which sows confusion as to which should be used for a given application. The following criteria can help you decide: ●

How many keys will you be sorting? - For small amounts of data (say ), it really doesn't matter much which of the quadratic-time algorithms you use. Insertion sort is faster, simpler, and less likely to be buggy than bubblesort. Shellsort is much faster than insertion sort, but it involves somewhat trickier programming and looking up good insert sequences in Knuth [Knu73b].

file:///E|/BOOK/BOOK4/NODE148.HTM (1 of 5) [19/1/2003 1:30:37]

Sorting

●

●

●

If you have more than 100 items to sort, it is important to use an -time algorithm, like heapsort, quicksort, or mergesort. If you have more than 1,000,000 items to sort, you probably need an external-memory algorithm that minimizes disk access. Both types of algorithms are discussed below. Will there be duplicate keys in the data? - When all items have distinct keys, the sorted order is completely defined. However, when two items share the same key, something else must determine which one comes first. For many applications it doesn't matter, so any sorting algorithm is equally good. Often, ties are broken by sorting on a secondary key, like the first name or initial if the family names collide. Occasionally, ties need to be broken by their initial position in the data set. If the 5th and 27th items of the initial data set share the same key in such a case, the 5th item must be before the 27th in the final order. A stable sorting algorithm preserves the original ordering in case of ties. Most of the quadratic-time sorting algorithms are stable, while many of the algorithms are not. If it is important that your sort be stable, it is probably better to explicitly use the initial position as a secondary key rather than trust the stability of your implementation. What do you know about your data? - In special applications, you can often exploit knowledge about your data to get it sorted faster or more easily. Of course, general sorting is a fast algorithm, so if the time spent sorting is really the bottleneck in your application, you are a fortunate person indeed. ❍ Is the data already partially sorted? If so, algorithms like insertion sort perform better than they otherwise would. ❍ Do you know the distribution of the keys? If the keys are randomly or uniformly distributed, a bucket or distribution sort makes sense. Throw the keys into bins based on their first letter, and recur until each bin is small enough to sort by brute force. This is very efficient when the keys get evenly distributed into buckets. However, bucket sort would be bad news sorting names on the mailing list of the ``Smith Society.'' ❍ Are your keys very long or hard to compare? If your keys are long text strings, it might pay to use a radix or bucket sort instead of a standard comparison sort, because the time of each comparison can get expensive. A radix sort always takes time linear in the number of characters in the file, instead of times the cost of comparing two keys. ❍ Is the range of possible keys very small? If you want to sort a subset of n/2 distinct integers, each with a value from 1 to n, the fastest algorithm would be to initialize an nelement bit vector, turn on the bits corresponding to keys, then scan from left to right and report which bits are on. Do I have to worry about disk accesses? - In massive sorting problems, it may not be possible to keep all data in memory simultaneously. Such a problem is called external sorting, because one must use an external storage device. Traditionally, this meant tape drives, and Knuth [Knu73b] describes a variety of intricate algorithms for efficiently merging data from different tapes. Today, it usually means virtual memory and swapping. Any sorting algorithm will work with virtual memory, but most will spend all their time swapping.

file:///E|/BOOK/BOOK4/NODE148.HTM (2 of 5) [19/1/2003 1:30:37]

Sorting

●

The simplest approach to external sorting loads the data into a B-tree (see Section ) and then does an in-order traversal of the tree to read the keys off in sorted order. Other approaches are based on mergesort. Files containing portions of the data are sorting using a fast internal sort, and then these files are merged in stages using 2- or k-way merging. Complicated merging patterns based on the properties of the external storage device can be used to optimize performance. How much time do you have to write and debug your routine? - If I had under an hour to deliver a working routine, I would probably just use a simple selection sort. If I had an afternoon to build an efficient sort routine, I would probably use heapsort, for it delivers reliable performance without tuning. If I was going to take the time required to build a fast system sort routine, I would carefully implement quicksort.

The best general-purpose sorting algorithm is quicksort (see Section ), although it requires considerable tuning effort to achieve maximum performance. Indeed, you are probably better off using a library function instead of doing it yourself. A poorly written quicksort will likely run more slowly than a poorly written heapsort. If you are determined to implement your own quicksort, use the following heuristics, which make a big difference in practice: ●

●

●

●

Use randomization - By randomly permuting (see Section ) the keys before sorting, you can eliminate the potential embarrassment of quadratic-time behavior on nearly-sorted data. Median of three - For your pivot element, use the median of the first, last, and middle elements of the array, to increase the likelihood of partitioning the array into roughly equal pieces. Some experiments suggest using a larger sample on big subarrays and a smaller sample on small ones. Leave small subarrays for insertion sort - Terminating the quicksort recursion and switching to insertion sort makes sense when the subarrays get small, say fewer than 20 elements. You should experiment to determine the best switchpoint for your implementation. Do the smaller partition first - Assuming that your compiler is smart enough to remove tail recursion, you can minimize runtime memory by processing the smaller partition before the larger one. Since successive stored calls are at most half as large as before, only stack space is needed.

Before you get started, see Bentley's article on building a faster quicksort [Ben92b]. Implementations: Pascal implementations of all the primary sorting algorithms are available from [MS91]. See Section the winner.

for details. Timing comparisons show an optimized version of quicksort to be

Bare bones implementations of all basic sorting algorithms, in C and Pascal, appear in [GBY91]. Most notable is the inclusion of implementations of external memory sorting algorithms. Sedgewick includes file:///E|/BOOK/BOOK4/NODE148.HTM (3 of 5) [19/1/2003 1:30:37]

Sorting

similar sort routine fragments in C++. See Section

for details.

) is an algorithm animation system for UNIX and X-windows, which includes XTango (see Section animations of all the basic sorting algorithms, including bubblesort, heapsort, mergesort, quicksort, radix sort, and shellsort. Many of these are quite interesting to watch. Indeed, sorting is the canonical problem for algorithm animation. Algorithm 410 [Cha71] and Algorithm 505 [Jan76] of the Collected Algorithms of the ACM are Fortran codes for sorting. The latter is an implementation of Shellsort on linked lists. Both are available from Netlib (see Section

).

C language implementations of Shellsort, quicksort, and heapsort appear in [BR95]. The code for these algorithms is printed in the text and available on disk for a modest fee. A bare bones implementation of heapsort in Fortran from [NW78] can be obtained in Section bare bones implementation of heapsort in Mathematica from [Ski90] can be obtained in Section

.

A .

Notes: Knuth [Knu73b] is the best book that has been written on sorting and indeed is the best book that will ever be written on sorting. It is now almost twenty-five years old, and a revised edition is promised, but it remains fascinating reading. One area that has developed since Knuth is sorting under presortedness measures. A newer and noteworthy reference on sorting is [GBY91], which includes pointers to algorithms for partially sorted data and includes implementations in C and Pascal for all of the fundamental algorithms. Expositions on the basic internal sorting algorithms appear in every algorithms text, including [AHU83, Baa88, CLR90, Man89]. Treatments of external sorting are rarer but include [AHU83]. Heapsort was first invented by Williams [Wil64]. Quicksort was invented by Hoare [Hoa62], with careful analysis and implementation by Sedgewick [Sed78]. Von Neumann is credited with having produced the first implementation of mergesort, on the EDVAC in 1945. See Knuth for a full discussion of the history of sorting, dating back to the days of punched-card tabulating machines. Sorting has a well-known lower bound under the algebraic decision tree model [BO83]. Determining the exact number of comparisons required for sorting n elements, for small values of n, has generated considerable study. See [Aig88, Raw92] for expositions. This lower-bound does not hold under different models of computation. Fredman and Willard [FW93] algorithm for sorting under a model of computation that permits arithmetic present an operations on keys. Under a similar model, Thorup [Tho96] developed a priority queue supporting file:///E|/BOOK/BOOK4/NODE148.HTM (4 of 5) [19/1/2003 1:30:37]

Sorting

operations, implying an Related Problems: Dictionaries (see page

sorting algorithm. ), searching (see page

), topological sorting (see page

).

Next: Searching Up: Combinatorial Problems Previous: Combinatorial Problems Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE148.HTM (5 of 5) [19/1/2003 1:30:37]

Searching

Next: Median and Selection Up: Combinatorial Problems Previous: Sorting

Searching

Input description: A set S of n keys, a query key q. Problem description: Where is q in S? Discussion: Searching means different things to different people. Searching for the global maximum or minimum of a function is the problem of unconstrained optimization and is discussed in Section . Chess playing programs search for the best move to make next by using alpha-beta minimax search, which is an exhaustive search of the possible moves using a variation of backtracking (see Section

).

Here we consider the simple task of searching for a key in a list or in an array, which is a fundamental problem associated with information retrieval. Dictionary data structures maintain efficient access to sets of keys under insertion and deletion and are discussed in Section binary trees and hash tables.

. Typical dictionaries include

We treat searching here as a problem distinct from dictionaries because simpler and more efficient solutions emerge when our primary interest is static searching. These little data structures can yield file:///E|/BOOK/BOOK4/NODE149.HTM (1 of 4) [19/1/2003 1:30:39]

Searching

large performance improvements when properly employed in an innermost loop. Also, several of the ideas from list and array searching, such as binary search and self-organization, apply to other problems and justify our attention. There are two basic approaches to array searching: sequential search and binary search. Both are simple, yet have interesting and subtle variations. In sequential search, we simply start from the front of our list or array of keys and compare each successive item against the key until we find a match or reach the end. In binary search, we start with a sorted array of keys. To search for key q, we compare q to the middle key . If q is before , it must reside in the top half of our set; if not, it must reside in the bottom half of our set. By repeating this process on the correct half, we find the key in a total of comparisons. This is a big win over the n/2 comparisons we expect with sequential search. See Section for more on binary search. Sequential search is the simplest algorithm, and likely to be fastest on up to 10-20 elements. For such tiny problems, forget about binary search. Beyond 50-100 elements, there is no question that binary search will be more efficient than sequential search, even factoring in the cost of the sorting (assuming multiple queries). Other issues do come into play, however, particularly in identifying the proper variant of the algorithm: ●

●

How much time can you spend programming? - Binary search is a notoriously tricky algorithm to program correctly. It took seventeen years after its invention until the first correct version of binary search was published! Don't be afraid to start from one of the implementations described below. Test it completely by writing a driver that searches for every key in the set S as well as between the keys. Are certain items accessed more often than other ones? - Certain English words (such as ``the'') are much more likely to occur than others (such as ``defenestrate''). We can reduce the number of comparisons in a sequential search by putting the most popular words at the top of the list and the least popular ones at the bottom. Further, nonuniform access is typically the rule, not the exception. Many real-world distributions, such as word use in English, are more accurately modeled by Zipf's law. Under Zipf's law, the ith most frequently accessed key is selected with probability (i-1)/i times the probability of the (i-1)st most popular key, for all . However, preordering the list to exploit a skewed access pattern requires knowing the access pattern in advance. For many applications, it can be difficult to obtain such information. Far easier are self-organizing lists, where the order of the keys changes in response to the queries. The simplest and best self-organizing scheme is move-to-front; that is, we move the most recently searched-for key from its current position to the front of the list. Popular keys keep getting boosted to the front, while unsearched-for keys drift towards the back of the list. There is no need to keep track of the frequency of access; just move the keys on demand. Self-organizing lists also exploit locality of reference, since accesses to a given key are likely to occur in clusters. Any key will be maintained near the top of the list during a cluster of accesses, even if other keys have proven more popular in the past.

file:///E|/BOOK/BOOK4/NODE149.HTM (2 of 4) [19/1/2003 1:30:39]

Searching

●

●

●

Self-organization can extend the useful size range of sequential search. However, you should switch to binary search beyond 50-100 elements. Is the key close by? - Suppose we know that the target key is to the right of position p, and we think it is close by. Sequential search is fast if we are correct, but we will be punished severely whenever we guess wrong. A better idea is to test repeatedly at larger intervals (p+1, p+2, p+4, p+8, p+16, ) to the right until we find a key to the right of our target. After this, we have a window containing the target and we can proceed with binary search. comparisons, Such a one-sided binary search finds the target at position p+l using at most so it is faster than binary search when l < < n, yet it can never be much worse. One-sided binary search is particularly useful in unbounded search problems, such as in numerical root finding. Is my data structure sitting on external memory? Once the number of keys grows too large, as in a CD-ROM telephone directory of all the people in the United States, binary search loses its status as the best search technique. Binary search jumps wildly around the set of keys looking for midpoints to compare, and it becomes very expensive to read in a new page from a secondary storage device for each comparison. Much better are data structures such as B-trees (see Section ), which cluster the keys into pages so as to minimize the number of disk accesses per search. Can I guess where the key should be? - In interpolation search, we exploit our understanding of the distribution of keys to guess where to look next. Interpolation search is probably a more accurate description of how we use a telephone book than binary search. For example, suppose we are searching for Washington, George in a sorted telephone book. We would certainly be safe making our first comparison three-fourths of the way down the list, essentially doing two comparisons for the price of one. Although interpolation search is an appealing idea, we caution against it for three reasons: First, you have to work very hard to optimize your search algorithm before you can hope for a speedup over binary search. Second, even if you get lucky and beat binary search, it is unlikely to be by enough to have justified the exercise. Third, your program will be much less robust and efficient when the distribution changes, such as when your application gets put to work on French words instead of English.

Implementations: The basic sequential and binary search algorithms are simple enough to implement that you should likely do them yourself. Still, the routines described below may be useful as models. Gonnet and Baeza-Yates provides code fragments in C and Pascal for sequential, binary, and ) provides a interpolation search, as well as for related dictionary structures. LEDA (see Section sorted array data type in C++ that supports binary search. Many textbooks include implementations of binary search, including [MS91]. See Section

file:///E|/BOOK/BOOK4/NODE149.HTM (3 of 4) [19/1/2003 1:30:39]

for details.

Searching

Notes: Mehlhorn and Tsakalidis [MT90b] give a through survey of the state-of-the-art in modern data structures. Knuth [Knu73a] provides a detailed analysis and exposition on all fundamental search algorithms and dictionary data structures but omits such modern data structures as red-black and splay trees. Gonnet and Baeza-Yates [GBY91] provide detailed references and experimental results for a wide variety of search algorithms. Manber [Man89] provides an interesting discussion of variations of binary search, including one-sided . binary search and searching for an index in A where In linear interpolation search on an array of sorted numbers, the next position probed is given by

where q is the query numerical key and S the sorted numerical array. If the keys are drawn independently from a uniform distribution, the expected search time is [YY76]. Expositions on interpolation search include [Raw92]. Nonuniform access patterns can be exploited in binary search trees by structuring them so that popular keys are located near the root, thus minimizing search time. Dynamic programming can be used to construct such optimal search trees in time [Knu73b]. Expositions include [AHU74]. Splay trees are self-organizing tree structures, as discussed in Section Related Problems: Dictionaries (see page

), sorting (see page

. ).

Next: Median and Selection Up: Combinatorial Problems Previous: Sorting Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE149.HTM (4 of 4) [19/1/2003 1:30:39]

Median and Selection

Next: Generating Permutations Up: Combinatorial Problems Previous: Searching

Median and Selection

Input description: A set of n numbers or keys, and an integer k. Problem description: Find the key that is smaller than exactly k of the n keys. Discussion: Median finding is an essential problem in statistics, where it provides a more robust notion of average than the mean. The mean wealth of people who have published research papers on sorting is significantly affected by the presence of one William Gates [GP79], although his effect on the median wealth is merely to cancel out one starving graduate student. Median finding is a special case of the more general selection problem, which asks for the ith element in sorted order. Selection arises in several applications: ●

●

Filtering outlying elements - In dealing with noisy data samples, it is usually a good idea to throw out the largest and smallest 10% or so of them. Selection can be used to identify the items defining the tenth and ninetieth percentiles, and the outliers are then filtered out by comparing each item to the two selected elements. Identifying the most promising candidates - In a computer chess program, we might quickly evaluate all possible next moves, and then decide to study the top 25% more carefully. Selection

file:///E|/BOOK/BOOK4/NODE150.HTM (1 of 3) [19/1/2003 1:30:40]

Median and Selection

●

followed by filtering is the way to go. Order statistics Particularly interesting special cases of selection include finding the smallest element (i=1), the largest element (i=n), and the median element (i= n/2).

The mean of n numbers can be easily computed in linear time by summing the elements and dividing by n. However, finding the median is a more difficult problem. Algorithms that compute the median can easily be generalized to arbitrary selection. The most elementary median-finding algorithm sorts the items in time and then returns the item sitting the (n/2)nd position. The good thing is that this gives much more information than just the median, enabling you to select the ith element (for all ) in constant time after the sort. However, there are faster algorithms if all you want is the median. In particular, there is an O(n) expected-time algorithm based on quicksort. Select a random element in the data set as a pivot, and use it to partition the data into sets of elements less than and greater than the pivot. From the sizes of these sets, we know the position of the pivot in the total order, and hence whether the median lies to the left or right of the pivot. Now we recur on the appropriate subset until it converges on the median. This takes (on average) iterations, with the cost of each iteration being roughly half that of the previous one. This defines a geometric series that converges to a linear-time algorithm, although if you are very unlucky it takes the same time as quicksort, . More complicated algorithms are known that find the median in worst-case linear time. However, the expected-time algorithm will likely win in practice. Just make sure to select random pivots in order to avoid the worst case. Beyond mean and median, a third notion of average is the mode, defined to be the element that occurs the greatest number of times in the data set. The best way to compute the mode sorts the set in time, which places all identical elements next to each other. By doing a linear sweep from left to right on this sorted set, we can count the length of the longest run of identical elements and hence compute the mode in a total of time. In fact, there is no faster worst-case algorithm possible to compute the mode, since the problem of testing whether there exist two identical elements in a set (called element uniqueness) can be shown to have an lower bound. Element uniqueness is equivalent to asking if the mode is . Possibilities exist, at least theoretically, for improvements when the mode is large by using fast median computations. Implementations: A bare bones implementation in C of the recursive k-selection algorithm appears in [GBY91]. See Section XTango (see Section

for further details. ) is an algorithm animation system for UNIX and X-windows, which includes

file:///E|/BOOK/BOOK4/NODE150.HTM (2 of 3) [19/1/2003 1:30:40]

Median and Selection

an animation of the linear-time selection algorithm. This animation is a good one. Notes: The linear expected-time algorithm for median and selection is due to Hoare [Hoa61]. Floyd and Rivest [FR75] provide an algorithm that uses fewer comparisons on average. Good expositions on lineartime selection include [AHU74, Baa88, CLR90, Raw92], with [Raw92] being particularly enlightening. A sport of considerable theoretical interest is determining exactly how many comparisons are sufficient to find the median of n items. The linear-time algorithm of Blum et. al. [BFP 72] proves that suffice, but we want to know what c is. A lower bower bound of 2n comparisons for median finding was given by Bent and John [BJ85]. In 1976, Schönhage, Paterson, and Pippenger [SPP76] presented an algorithm using 3n comparisons. Recently, Dor and Zwick [DZ95] proved that 2.95 n comparisons suffice. These algorithms attempt to minimize the number of element comparisons but not the total number of operations, and hence do not lead to faster algorithms in practice. Tight combinatorial bounds for selection problems are presented in [Aig88]. An optimal algorithm for computing the mode is presented in [DM80]. Related Problems: Priority queues (see page

), sorting (see page

).

Next: Generating Permutations Up: Combinatorial Problems Previous: Searching Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE150.HTM (3 of 3) [19/1/2003 1:30:40]

Generating Permutations

Next: Generating Subsets Up: Combinatorial Problems Previous: Median and Selection

Generating Permutations

Input description: An integer n. Problem description: Generate (1) all or (2) a random or (3) the next permutation of length n. Discussion: A permutation describes an arrangement, or ordering, of things. Many algorithmic problems in this catalog seek the best way to order a set of objects, including traveling salesman (the least-cost order to visit n cities), bandwidth (order the vertices of a graph on a line so as to minimize the length of the longest edge), and graph isomorphism (order the vertices of one graph so that it is identical to another). Any algorithm for solving such problems exactly must construct a series of permutations along the way. There are n! permutations of n items, which grows so quickly that you can't expect to generate all permutations for n > 11, since 11! = 39,916,800. Numbers like these should cool the ardor of anyone interested in exhaustive search and help explain the importance of generating random permutations. Fundamental to any permutation-generation algorithm is a notion of order, the sequence in which the permutations are constructed, from first to last. The most natural generation order is lexicographic, the order they would appear if they were sorted numerically. Lexicographic order for n=3 is , file:///E|/BOOK/BOOK4/NODE151.HTM (1 of 5) [19/1/2003 1:30:42]

Generating Permutations

, , , , and finally . Although lexicographic order is aesthetically pleasing, there is often no particular reason to use it. For example, if you are searching through a collection of files, it does not matter whether the filenames are encountered in sorted order, so long as you search through all of them. Indeed, nonlexicographic orders lead to faster and simpler permutation generation algorithms. There are two different paradigms for constructing permutations: ranking/unranking and incremental change methods. Although the latter are more efficient, ranking and unranking can be applied to solve a much wider class of problems, including the other combinatorial generation problems in this book. The key is to define functions rank and unrank on all permutations p and integers n, m, where |p| = n and . ●

●

Rank(p) - What is the position of p in the given generation order? A typical ranking function is recursive, such as , with . Getting this right means relabeling the elements of the smaller permutation to reflect the deleted first element. Thus

Unrank(m,n) - Which permutation is in position m of the n! permutations of n items? A typical unranking function finds the number of times (n-1)! goes into m and proceeds recursively. Unrank(2,3) tells us that the first element of the permutation must be `2', since but . Deleting from m leaves the smaller problem Unrank(0,2). The ranking of 0 corresponds to the total order, and the total order on the two remaining elements (since `2' has been used) is , so .

What the actual rank and unrank functions are does not matter as much as the fact that they must be inverses. In other words, p = Unrank(Rank(p), n) for all permutations p. Once you define ranking and unranking functions for permutations, you can solve a host of related problems: ●

● ●

Sequencing permutations - To determine the next permutation that occurs in order after p, we can Rank(p), add 1, and then Unrank(p). Similarly, the permutation right before p in order is Unrank(Rank(p)-1,|p|). Counting through the integers from 0 to n!-1 and unranking them is equivalent to generating all permutations. Generating random permutations - Select a random integer from 0 to n!-1 and unrank it. Keep track of a set of permutations - Suppose we want to construct random permutations and act only when we encounter one we have not seen before. We can set up a bit vector (see Section ) with n! bits, and set bit i to 1 if permutation Unrank(i,n) has been seen. A similar technique was employed with k-subsets in the Lotto application of Section

file:///E|/BOOK/BOOK4/NODE151.HTM (2 of 5) [19/1/2003 1:30:42]

.

Generating Permutations

The rank/unrank method is best suited for small values of n, since n! quickly exceeds the capacity of ). The incremental machine integers, unless arbitrary-precision arithmetic is available (see Section change methods work by defining the next and previous operations to transform one permutation into another, typically by swapping two elements. The tricky part is to schedule the swaps so that permutations do not repeat until all of them have been generated. See the output picture above for an ordering of the six permutations of with a single swap between successive permutations. Incremental change algorithms for sequencing permutations are tricky, but they are concise enough that they can be expressed in a dozen lines of code. See the implementation section for pointers to code. Because the incremental change is a single swap, these algorithms can be extremely fast - on average, constant time - which is independent of the size of the permutation! The secret is to represent the permutation using an n-element array to facilitate the swap. In certain applications, only the change between permutations is important. For example, in a brute-force program to search for the optimal tour, the cost of the tour associated with the new permutation will be that of the previous permutation, with the addition and deletion of four edges. Throughout this discussion, we have assumed that the items we are permuting are all distinguishable. However, if there are duplicates (meaning our set is a multiset), you can save considerable time and effort by avoiding identical permutations. For example, there are only ten permutations of , instead of 120. To avoid duplicates use backtracking and generate the permutations in lexicographic order. Generating random permutations is an important little problem that people stumble across often, and often botch up. The right way is the following two-line, linear-time algorithm. We assume that Random[i,n] generates a random integer between i and n, inclusive.

for i=1 to n do a[i] = i;

for i=1 to n-1 do swap[ a[i], a[ Random[i,n] ];

That this algorithm generates all permutations uniformly at random is not obvious. If you think so, explain convincingly why the following algorithm does not generate permutations uniformly:

for i=1 to n do a[i] = i;

for i=1 to n-1 do swap[ a[i], a[ Random[1,n] ]; file:///E|/BOOK/BOOK4/NODE151.HTM (3 of 5) [19/1/2003 1:30:42]

Generating Permutations

Such subtleties demonstrate why you must be very careful with random generation algorithms. Indeed, we recommend that you try some reasonably extensive experiments with any random generator before really believing it. For example, generate 10,000 random permutations of length 4 and see whether all 24 of them occur approximately the same number of times. If you understand how to measure statistical significance, you are in even better shape. Implementations: The best source on generating combinatorial objects is Nijenhuis and Wilf [NW78], who provide efficient Fortran implementations of algorithms to construct random permutations and to sequence permutations in minimum-change order. Also included are routines to extract the cycle structure of a permutation. See Section

for details.

An exciting WWW site developed by Frank Ruskey of the University of Victoria contains a wealth of material on generating combinatorial objects of different types, including permutations, subsets, partitions, and certain graphs. Specifically, there is an interactive interface that lets you specify which type of objects you would like to construct and quickly returns the objects to you. It is well worth checking this out at http://sue.csc.uvic.ca/ cos/. Combinatorica [Ski90] provides Mathematica implementations of algorithms that construct random permutations and sequence permutations in minimum change and lexicographic orders. It also provides a backracking routine to construct all distinct permutations of a multiset, and it supports various permutation group operations. See Section The Stanford GraphBase (see Section

. ) contains routines to generate all permutations of a multiset.

Notes: The primary reference on permutation generation is the survey paper by Sedgewick [Sed77]. Good expositions include [NW78, RND77, Rus97]. The fast permutation generation methods make only a single swap between successive permutations. The Johnson-Trotter algorithm [Joh63, Tro62] satisfies an even stronger condition, namely that the two elements being swapped are always adjacent. In the days before ready access to computers, books with tables of random permutations [MO63] were used instead of algorithms. The swap-based random permutation algorithm presented above was first described in [MO63]. Related Problems: Random number generation (see page

file:///E|/BOOK/BOOK4/NODE151.HTM (4 of 5) [19/1/2003 1:30:42]

), generating subsets (see page

),

Generating Permutations

generating partitions (see page

).

Next: Generating Subsets Up: Combinatorial Problems Previous: Median and Selection Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE151.HTM (5 of 5) [19/1/2003 1:30:42]

Generating Subsets

Next: Generating Partitions Up: Combinatorial Problems Previous: Generating Permutations

Generating Subsets

Input description: An integer n. Problem description: Generate (1) all or (2) a random or (3) the next subset of the integers

.

Discussion: A subset describes a selection of objects, where the order among them does not matter. Many of the algorithmic problems in this catalog seek the best subset of a group of things: vertex cover seeks the smallest subset of vertices to touch each edge in a graph; knapsack seeks the most profitable subset of items of bounded total size; and set packing seeks the smallest subset of subsets that together cover each item exactly once. There are distinct subsets of an n-element set, including the empty set as well as the set itself. This grows exponentially, but at a considerably smaller rate than the n! permutations of n items. Indeed, since 1,048,576, a brute-force search through all subsets of 20 elements is easily manageable, although by n=30, 1,073,741,824, so you will certainly be pushing things. By definition, the relative order among the elements does not distinguish different subsets. Thus . However, it is a very good idea to maintain your subsets in a sorted or canonical is the same as order, in order to speed up such operations as testing whether two subsets are identical or making them file:///E|/BOOK/BOOK4/NODE152.HTM (1 of 4) [19/1/2003 1:30:44]

Generating Subsets

look right when printed. As with permutations (see Section ), the key to subset generation problems is establishing a numerical sequence among all subsets. There are three primary alternatives: ●

●

●

Lexicographic order - Lexicographic order is sorted order, and often the most natural order for generating combinatorial objects. The eight subsets of in lexicographic order are , , , , , , , and . Unfortunately, it is surprisingly difficult to generate subsets in lexicographic order. Unless you have a compelling reason to do so, forget about it. Gray Code - A particularly interesting and useful subset sequence is the minimum change order, wherein adjacent subsets differ by the insertion or deletion of exactly one element. Such an ordering, called a Gray code, appears in the output picture above. Subset generation in Gray code order can be very fast, because there is a nice recursive construction to sequence subsets. Further, since only one element changes between subsets, exhaustive search algorithms built on Gray codes can be quite efficient. A set cover program would only have to update the change in coverage by the addition or deletion of one subset. See the implementation section below for Gray code subset generation programs. Binary counting - The simplest approach to subset generation problems is based on the observation that any subset S' of S is defined by which of the n=|S| items are in S'. We can represent S' by a binary string of n bits, where bit i is 1 iff the ith element of S is in S'. This defines a bijection between the binary strings of length n, and the subsets of n items. For n=3, binary counting generates subsets in the following order: {}, {3}, {2}, {2,3}, {1}, {1,3}, {1,2}, {1,2,3}. This binary representation is the key to solving all subset generation problems. To generate all subsets in order, simply count from 0 to . For each integer, successively mask off each of the bits and compose a subset of exactly the items corresponding to `1' bits. To generate the next or previous subset, increment or decrement the integer by one. Unranking a subset is exactly the masking procedure, while ranking constructs a binary number with 1's corresponding to items in S and then converts this binary number to an integer. To generate a random subset, you could generate a random integer from 0 to and unrank, although you are probably asking for trouble because any flakiness with how your random number generator rounds things off means that certain subsets can never occur. Therefore, a better approach is simply to flip a coin n times, with the ith flip deciding whether to include element i in the subset. A coin flip can be robustly simulated by generating a random real or large integer and testing whether it is bigger or smaller than half the range. A Boolean array of n items can thus be used to represent subsets as a sort of premasked integer. The only complication is that you must explicitly handle the carry if you seek to generate all subsets.

file:///E|/BOOK/BOOK4/NODE152.HTM (2 of 4) [19/1/2003 1:30:44]

Generating Subsets

Generation problems for two closely related problems arise often in practice: ●

k-subsets - Instead of constructing all subsets, we may only be interested in the subsets containing exactly k elements. There are such subsets, which is substantially less than

,

particularly for small values of k. The best way to construct all k-subsets is in lexicographic order. The ranking function is based on the observation that there are k-subsets whose smallest element is f. Using this, it is

●

possible to determine the smallest element in the mth k-subset of n items. We then proceed recursively for subsequent elements of the subset. See the implementations below for details. Strings - Generating all subsets is equivalent to generating all strings of true and false. To generate all or random strings on alphabets of size , the same basic techniques apply, except there will be strings in total.

Implementations: Nijenhuis and Wilf [NW78] provide efficient Fortran implementations of algorithms to construct random subsets and to sequence subsets in Gray code and lexicographic order. They also provide routines to construct random k-subsets and sequence them in lexicographic order. See Section for details on ftp-ing these programs. Algorithm 515 [BL77] of the Collected Algorithms of the ACM is another Fortran implementation of lexicographic k-subsets, available from Netlib

(see Section

).

An exciting WWW site developed by Frank Ruskey of the University of Victoria contains a wealth of material on generating combinatorial objects of different types, including permutations, subsets, partitions, and certain graphs. Specifically, it provides an interactive interface that lets you specify which type of objects you would like to construct and then returns the objects to you. Check this out at http://sue.csc.uvic.ca/ cos/. Combinatorica [Ski90] provides Mathematica implementations of algorithms to construct random subsets and to sequence subsets in Gray code, binary, and lexicographic order. They also provide routines to construct random k-subsets and strings, and sequence them lexicographically. See Section for further information on Combinatorica. Notes: The primary expositions on subset generation include [NW78, RND77, Rus97]. Wilf [Wil89] provides an update of [NW78], including a thorough discussion of modern Gray code generation problems. Gray codes were first developed [Gra53] to transmit digital information in a robust manner over an analog channel. By assigning the code words in Gray code order, the ith word differs only slightly from the (i+1)st, so minor fluctuations in analog signal strength corrupts only a few bits. Gray codes have a particularly nice correspondence to Hamiltonian cycles on the hypercube. See any of the references file:///E|/BOOK/BOOK4/NODE152.HTM (3 of 4) [19/1/2003 1:30:44]

Generating Subsets

above for details. An exposition on the more general problem of constructing Gray codes for k items (instead of subsets) appears in [Man89]. The popular puzzle Spinout, manufactured by Binary Arts Corporation, can be solved using Gray codes. Related Problems: Generating permutations (see page

), generating partitions (see page

Next: Generating Partitions Up: Combinatorial Problems Previous: Generating Permutations Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE152.HTM (4 of 4) [19/1/2003 1:30:44]

).

Generating Partitions

Next: Generating Graphs Up: Combinatorial Problems Previous: Generating Subsets

Generating Partitions

Input description: An integer n. Problem description: Generate (1) all or (2) a random or (3) the next integer or set partitions of length n. There are two different types of combinatorial objects denoted by the term ``partition'', namely integer partitions and set partitions. Although they are quite different beasts, it is a good idea to make both a part of your vocabulary: ●

●

Integer partitions of n are sets of nonzero integers that add up to exactly n. For example, the seven distinct integer partitions of 5 are {5}, {4,1}, {3,2}, {3,1,1}, {2,2,1}, {2,1,1,1}, and {1,1,1,1,1}. An interesting application I encountered that required the generation of integer partitions was in a simulation of nuclear fission. When an atom is smashed, the nucleus of protons and neutrons is broken into a set of smaller clusters. The sum of the particles in the set of clusters must equal the original size of the nucleus. As such, the integer partitions of this original size represent all the possible ways to smash the atom. Set partitions divide the elements into nonempty subsets. For example, there are fifteen distinct set partitions of n=4: {1234}, {123,4}, {124,3}, {12,34}, {12,3,4}, {134,2}, {13,24}, {13,2,4}, {14,23}, {1,234}, {1,23,4}, {14,2,3}, {1,24,3}, {1,2,34}, and {1,2,3,4}. Several of the problems in this catalog return set partitions as results, such as vertex coloring and connected components.

file:///E|/BOOK/BOOK4/NODE153.HTM (1 of 4) [19/1/2003 1:30:46]

Generating Partitions

Although the number of integer partitions grows exponentially with n, they do so at a refreshingly slow rate. There are only 627 partitions of n=20, and it is even possible to enumerate all partitions of n=100, since there there are only 190,569,292 of them. The best way to generate all partitions is to construct them in lexicographically decreasing order. The first partition is itself. The general rule is to subtract 1 from the smallest part that is >1 and then collect all the 1's so as to match the new smallest part >1. For example, the partition after is , since the five 1's left after 3-1=2 becomes the smallest part are best packaged as 2,2,1. When the partition is all 1's, we have completed one trip through all the partitions. This algorithm is not particularly complicated, but it is sufficiently intricate that you should consider using one of the implementations below. In either case, test it to make sure that you get exactly 627 distinct partitions for n=20. Generating integer partitions uniformly at random is a trickier matter than generating random permutations or subsets. This is because selecting the first (i.e. largest) element of the partition has a dramatic effect on the number of possible partitions that can be generated. Observe that no matter how large n is, there is only one partition of n whose largest part is 1. The number of partitions of n with largest part at most k is given by the recurrence

with the two boundary conditions and . This function can be used to select the largest part of your random partition with the correct probabilities and, by proceeding recursively, to eventually construct the entire random partition. Implementations are cited below.

Figure: The Ferrers diagram of a random partition of n=1000 Random partitions tend to have large numbers of fairly small parts, best visualized by a Ferrers diagram . Each row of the diagram corresponds to one part of the partition, with the size of each as in Figure part represented by that many dots. file:///E|/BOOK/BOOK4/NODE153.HTM (2 of 4) [19/1/2003 1:30:46]

Generating Partitions

Set partitions can be generated using techniques similar to integer partitions. Each set partition is , where and , for encoded as a restricted growth function, . Each distinct digit identifies a subset, or block, of the partition, while the growth condition ensures that the blocks are sorted into a canonical order based on the smallest element in each block. For example, the restricted growth function 0,1,1,2,0,3,1 defines the set partition {{1,5}, {2,3,7}, {4}, {6} }. Since there is a one-to-one equivalence between set partitions and restricted growth functions, we can use lexicographic order on the restricted growth functions to order the set partitions. Indeed, the fifteen partitions of {1,2,3,4} listed above are sequenced according to the lexicographic order of their restricted growth function (check it out). To randomly generate set partitions, we use a similar counting strategy as with integer partitions. The Stirling numbers of the second kind S(n,k) count the number of partitions of 1,...,n with exactly k blocks. They are computed using the recurrence S(n,k) = S(n-1,k-1) + k*S(n-1,k) with the boundary conditions S(n,n)=1. The reader is referred to the references and implementations for more details and code. Implementations: The best source on generating combinatorial objects is Nijenhuis and Wilf [NW78], who provide efficient Fortran implementations of algorithms to construct random and sequential integer partitions, set partitions, compositions, and Young tableaux. See Section programs.

for details on ftp-ing these

An exciting WWW site developed by Frank Ruskey of the University of Victoria contains a wealth of material on generating combinatorial objects of different types, including permutations, subsets, partitions, and certain graphs. It is well worth checking this out at http://sue.csc.uvic.ca/~cos/. Combinatorica [Ski90] provides Mathematica implementations of algorithms to construct random and sequential integer partitions, compositions, strings, and Young tableaux, as well as to count and manipulate these objects. See Section

.

Algorithm 403 [BL77] of the Collected Algorithms of the ACM is a Fortran code for constructing integer partitions with k parts. It is available from Netlib Section

(see Section

). The Stanford GraphBase (see

) also contains generators for constructing all integer partitions.

Notes: The standard references on combinatorial generation [NW78, RND77, Rus97] all present algorithms for generating integer and/or set partitions. Andrews is the primary reference on integer partitions and related topics. Interestingly, the set of all 52 set partitions for n=5 appears in the form of Murasaki diagrams in the oldest novel known, The Tale of Genji. file:///E|/BOOK/BOOK4/NODE153.HTM (3 of 4) [19/1/2003 1:30:46]

Generating Partitions

Two related combinatorial objects are Young tableaux and integer compositions, although they are less likely to emerge in applications. Generation algorithms for both are presented in [NW78, RND77, Rus97] Young tableaux are two-dimensional configurations of integers {1,...,n} where the number of elements in each row is defined by an integer partition of n. Further, the elements of each row and column are sorted in increasing order, and the rows are left-justified. This notion of shape captures a wide variety of structures as special cases. They have many interesting properties, including the existance of a bijection between pairs of tableaux and permutations. Compositions represent the set of possible assignments of a set of n indistinguishable balls to k distinguishable boxes. For example, we can place three balls into two boxes as {3,0}, {2,1}, {1,2}, or {0,3}. Compositions are most easily constructed sequentially in lexicographic order. To construct them randomly, pick a random (k-1)-subset of n+k-1 items using the algorithm of the previous section, and count the number of unselected items between the selected ones. For example, if k=5 and n=10, the (5-1)subset {1,3,7,14} of 1,..., (n+k-1)=14 defines the composition {0,1,3,6,0}, since there are no items to the left of element 1 nor right of element 14. Related Problems: Generating permutations (see page

), generating subsets (see page

Next: Generating Graphs Up: Combinatorial Problems Previous: Generating Subsets Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE153.HTM (4 of 4) [19/1/2003 1:30:46]

).

Generating Graphs

Next: Calendrical Calculations Up: Combinatorial Problems Previous: Generating Partitions

Generating Graphs

Input description: Parameters describing the desired graph, such as the number of vertices n, the number of edges m, or the edge probability p. Problem description: Generate (1) all or (2) a random or (3) the next graph satisfying the parameters. Discussion: Graph generation typically arises in constructing test data for programs. Perhaps you have two different programs that solve the same problem, and you want to see which one is faster or make sure that they always give the same answer. Another application is experimental graph theory, verifying whether a particular property is true for all graphs or how often it is true. It is much easier to conjecture the four-color theorem once you have demonstrated 4-colorings for all planar graphs on 15 vertices. A different application of graph generation arises in network design. Suppose you need to design a network linking ten machines using as few cables as possible, such that the network can survive up to two vertex failures. One approach is to test all the networks with a given number of edges until you find one that will work. For larger graphs, more heuristic approaches, like simulated annealing, will likely be necessary. Many factors complicate the problem of generating graphs. First, make sure you know what you want to file:///E|/BOOK/BOOK4/NODE154.HTM (1 of 5) [19/1/2003 1:30:48]

Generating Graphs

generate: ●

●

Do I want labeled or unlabeled graphs? - The issue here is whether the names of the vertices matter in deciding whether two graphs are the same. In generating labeled graphs, we seek to construct all possible labelings of all possible graph topologies. In generating unlabeled graphs, we seek only one representative for each topology and ignore labelings. For example, there are only two connected unlabeled graphs on three vertices - a triangle and a simple path. However, there are four connected labeled graphs on three vertices - one triangle and three 3-vertex paths, each distinguished by their central vertex. In general, labeled graphs are much easier to generate. However, there are so many more of them that you quickly get swamped with isomorphic copies of the same few graphs. What do I mean by random? - There are two primary models of random graphs, both of which generate graphs according to different probability distributions. The first model is parameterized by a given edge probability p. Typically, p=0.5, although smaller values can be used to construct sparser random graphs. In this model a coin is flipped for each pair of vertices x and y to decide whether to add an edge (x,y). All labeled graphs will be generated with equal probability when p=1/2. The second model is parameterized by the desired number of edges m. It selects m distinct edges uniformly at random. One way to do this is by drawing random (x,y)-pairs and creating an edge if that pair is not already in the graph. An alternative approach to computing the same things constructs the set of possible edges and selects a random m-subset of them, as discussed in Section

.

Which of these options best models your application? Probably none of them. Random graphs, by definition, have very little structure. In most applications, graphs are used to model relationships, which are often highly structured. Experiments conducted on random graphs, although interesting and easy to perform, often fail to capture what you are looking for. An alternative to random graphs is to use ``organic'' graphs, graphs that reflect the relationships among real-world objects. The Stanford GraphBase, discussed below, is an outstanding source of organic graphs. Further, there are many raw sources of relationships electronically available via the Internet that can be turned into interesting organic graphs with a little programming and imagination. Consider the graph defined by a set of WWW pages, with any hyperlink between two pages defining an edge. Or what about the graph implicit in railroad, subway, or airline networks, with vertices being stations and edges between two stations connected by direct service? As a final example, every large computer program defines a call graph, where the vertices represent subroutines, and there is an edge (x,y) if x calls y. Two special classes of graphs have generation algorithms that have proven particularly useful in practice:

file:///E|/BOOK/BOOK4/NODE154.HTM (2 of 5) [19/1/2003 1:30:48]

Generating Graphs ●

Trees -

Prüfer codes provide a simple way to rank and unrank labeled trees and thus solve all

. There are exactly the standard generation problems discussed in Section on n vertices, and exactly that many strings of length n-2 on the alphabet

labeled trees .

The key to Prüfer's bijection is the observation that every tree has at least two vertices of degree 1. Thus in any labeled tree, the vertex v incident on the leaf with lowest label is well-defined. We take v to be , the first character in the code. We then delete the associated leaf and repeat the procedure until only two vertices are left. This defines a unique code S for any given labeled tree that can be used to rank the tree. To go from code to tree, observe that the degree of vertex v in the tree is one more than the number of times v occurs in S. The lowest-labeled leaf will be the smallest integer missing from S, which when paired with determines the first edge of the tree. The entire tree follows by induction.

●

Algorithms for efficiently generating unlabeled rooted trees are presented in the implementation section below. Fixed degree sequence graphs - The degree sequence of a graph G is an integer partition where is the degree of the ith highest-degree vertex of G. Since each edge contributes to the degree of two vertices, p is a partition of 2m, where m is the number of edges in G. Not all partitions correspond to degree sequences of graphs. However, there is a recursive construction that constructs a graph with a given degree sequence if one exists. If a partition is realizable, the highest-degree vertex can be connected to the next highest-degree vertices in G, or the vertices corresponding to parts . Deleting and decrementing yields a smaller partition, which we recur on. If we terminate without ever creating negative numbers, the partition was realizable. Since we always connect the highest-degree vertex to other high-degree vertices, it is important to reorder the parts of the partition by size after each iteration. Although this construction is deterministic, a semirandom collection of graphs realizing this degree sequence can be generated from G using edge-flipping operations. Suppose edges (x,y) and (w,z) are in G, but (x,w) and (y,z) are not. Exchanging these pairs of edges creates a different (not necessarily connected) graph without changing the degrees of any vertex.

Implementations: The Stanford GraphBase [Knu94] is perhaps most useful as an instance generator for constructing a wide variety of graphs to serve as test data for other programs. It incorporates graphs derived from interactions of characters in famous novels, Roget's Thesaurus, the Mona Lisa, expander graphs, and the economy of the United States. It also contains routines for generating binary trees, graph products, line graphs, and other operations on basic graphs. Finally, because of its machine-independent random number generators, it provides a way to construct random graphs such that they can be reconstructed elsewhere, thus making them perfect for experimental comparisons of algorithms. See Section

for additional information.

file:///E|/BOOK/BOOK4/NODE154.HTM (3 of 5) [19/1/2003 1:30:48]

Generating Graphs

Combinatorica [Ski90] provides Mathematica generators for basic graphs such as stars, wheels, complete graphs, random graphs and trees, and graphs with a given degree sequence. Further, it includes operations to construct more interesting graphs from these, including join, product, and line graph. Graffiti [Faj87], a collection of almost 200 graphs of graph-theoretic interest, are available in Combinatorica format. See Section

.

), by Brendan D. McKay of the The graph isomorphism testing program nauty (see Section Australian National University, has been used to generate catalogs of all nonisomorphic graphs with up to 11 vertices. This extension to nauty, named makeg, can be obtained by anonymous ftp from bellatrix.anu.edu.au (150.203.23.14) in the directory pub/nauty19. Nijenhuis and Wilf [NW78] provide efficient Fortran routines to enumerate all labeled trees via Prüfer codes and to construct random unlabeled rooted trees. See Section standard families of graphs is included with LEDA (see Section

.

A collection of generators for

).

Notes: An extensive literature exists on generating graphs uniformly at random. Surveys include [Gol93, Tin90]. Closely related to the problem of generating classes of graphs is counting them. Harary and Palmer [HP73] survey results in graphical enumeration. Random graph theory is concerned with the properties of random graphs. Threshold laws in random graph theory define the edge density at which properties such as connectedness become highly likely to occur. Expositions on random graph theory include [ES74, Pal85]. An integer partition is graphic if there exists a simple graph with that degree sequence. Erd•s and Gallai [EG60] proved that a degree sequence is graphic if and only if the sequence observes the following condition for each integer r < n:

The bijection between n-2 strings and labeled trees is due to Prüfer [Prü18]. Good expositions on this result include [Eve79a, NW78]. Related Problems: Generating permutations (see page

file:///E|/BOOK/BOOK4/NODE154.HTM (4 of 5) [19/1/2003 1:30:48]

), graph isomorphism (see page

).

Generating Graphs

Next: Calendrical Calculations Up: Combinatorial Problems Previous: Generating Partitions Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE154.HTM (5 of 5) [19/1/2003 1:30:48]

Calendrical Calculations

Next: Job Scheduling Up: Combinatorial Problems Previous: Generating Graphs

Calendrical Calculations

Input description: A particular calendar date d, specified by month, day, and year. Problem description: Which day of the week did d fall on according to the given calendar system? Discussion: Many business applications need to perform calendrical calculations. Perhaps we want to display a calendar of a specified month and year. Maybe we need to compute what day of the week or year some event occurs, as in figuring out the date on which a 180-day futures contract comes due. The importance of correct calendrical calculations is perhaps best revealed by the furor over the ``millennium bug,'' the crisis in legacy programs that allocate only two digits for storing the year. More complicated questions arise in international applications, because different nations and ethnic groups around the world use different calendar systems. Some of these, like the Gregorian calendar used in most of the world, are based on the sun, while others, like the Hebrew calendar, are lunar calendars. How would you tell today's date according to the Chinese or Arabic calendar? The algorithms associated with calendrical calculations are different from the other problems in this book, because calendars are historical objects, not mathematical ones. The issues revolve around specifying the rules of the calendrical system and simply implementing them correctly, rather than file:///E|/BOOK/BOOK4/NODE155.HTM (1 of 3) [19/1/2003 1:30:49]

Calendrical Calculations

designing efficient shortcuts for the computation. The basic approach behind calendar systems is to start with a particular reference date and count from there. The particular rules for wrapping the count around into months and years is what distinguishes a given calendar system from another. To implement a calendar, we need two functions, one that given a date returns the integer number of days that have elapsed since the reference start date, the other of which takes an integer n and returns the calendar date exactly n days from the reference date. This is analogous to the ranking and unranking rules for combinatorial objects, such as permutations (see Section

).

The major source of complications in calendar systems is that the solar year is not an integer number of days long. Thus if a calendar seeks to keep its annual dates in sync with the seasons, leap days must be added at both regular and irregular intervals. Since a solar year is 365 days and 5:49:12 hours long, an extra 10:48 minutes would have to be accounted for at the end of each year if we were simply to add a leap day every four years. The original Julian calendar (from Julius Caesar) did not account for these extra minutes, which had accumulated to ten days by 1582 when Pope Gregory XIII proposed the Gregorian calendar used today. Gregory deleted the ten days and eliminated leap days in years that are multiples of 100 but not 400. Supposedly, riots ensued because the masses feared their lives were being shortened by ten days. Outside the Catholic church, resistance to change slowed the reforms. The deletion of days did not occur in England and America until September 1752, and not until 1927 in Turkey. The rules for most calendrical systems are sufficiently complicated and pointless that you should lift code from a reliable place rather than attempt to write your own. We identify suitable implementations below. There are a variety of ``impress your friends'' algorithms that enable you to compute in your head on which day of the week a particular date occurred. Such algorithms often fail to work reliably outside the given century and should be avoided for computer implementation. Implementations: Dershowitz and Reingold provide a uniform algorithmic presentation [DR90, RDC93] for a variety of different calendar systems, including the Gregorian, ISO, Julian, Islamic, and Hebrew calendars, as well as other calendars of historical interest. Further, they provide Common Lisp and C++ routines to convert dates between calendars, day of the week computations, and the determination of secular and religious holidays. These are likely to be the most comprehensive and reliable calendrical routines you will be able to get your hands on, and are available at http://emr.cs.uiuc.edu:80/reingold/calendars.html. A nice package for calendrical computations in Mathematica by Ilan Vardi is available in the Packages/Miscellaneous directory of the standard Mathematica distribution, and also from MathSource. Vardi's book [Var91] discusses the theory behind the implementation, which provides support for the

file:///E|/BOOK/BOOK4/NODE155.HTM (2 of 3) [19/1/2003 1:30:49]

Calendrical Calculations

Gregorian, Julian, and Islamic calendars. Gregorian calendar computations implemented in C appear in [BR95]. This code uses 1582 as the date of the calendar reform, instead of the standard UNIX date of 1752. The code for these algorithms is printed in the text and are available on disk for a modest fee. Notes: The most comprehensive discussion of calendrical computation algorithms are the papers by Dershowitz and Reingold [DR90, RDC93]. These papers are superseded by their book [DR97]. Histories of the Gregorian calendar appear in [BR95]. Related Problems: Arbitrary-precision arithmetic (see page

), generating permutations (see page

).

Next: Job Scheduling Up: Combinatorial Problems Previous: Generating Graphs Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE155.HTM (3 of 3) [19/1/2003 1:30:49]

Job Scheduling

Next: Satisfiability Up: Combinatorial Problems Previous: Calendrical Calculations

Job Scheduling

Input description: A directed acyclic graph G=(V,E), where the vertices represent jobs and the edge (u,v) implies that task u must be completed before task v. Problem description: What schedule of tasks completes the job using the minimum amount of time or processors? Discussion: Devising a proper schedule to satisfy a set of constraints is fundamental to many applications. A critical aspect of any parallel processing system is the algorithm mapping tasks to processors. Poor scheduling can leave most of the expensive machine sitting idle while one bottleneck task is performed. Assigning people to jobs, meetings to rooms, or courses to final exam periods are all different examples of scheduling problems. Scheduling problems differ widely in the nature of the constraints that must be satisfied and the type of schedule desired. For this reason, several other catalog problems have a direct application to various kinds of scheduling: ●

To construct a schedule consistent with the precedence constraints, see topological sorting in Section

.

file:///E|/BOOK/BOOK4/NODE156.HTM (1 of 4) [19/1/2003 1:30:50]

Job Scheduling ●

To assign a set of jobs to people who have the appropriate skills for them, see bipartite matching

●

. in Section To assign a set of jobs to time slots such that no two jobs that interfere are assigned the same time

●

and . slot, see vertex and edge coloring in Sections To construct the most efficient route for a delivery person to visit a given set of locations, see the . To construct the most efficient route for a snowplow traveling salesman problem in Section or mailman to completely traverse a given set of edges, see the Eulerian cycle problem in Section .

In this section, we focus on precedence-constrained scheduling problems for directed acyclic graphs. These problems are often called PERT/CPM, for Program Evaluation and Review Technique/Critical Path Method. Suppose you have broken a big job into a large number of smaller tasks. For each task you know how long it should take (or perhaps an upper bound on how long it might take). Further, for each pair of tasks you know whether it is essential that one task be performed before another. The fewer constraints we have to enforce, the better our schedule can be. These constraints must define a directed acyclic graph, acyclic because a cycle in the precedence constraints represents a Catch-22 situation that can never be resolved. We are interested in several problems on these networks: ●

●

●

Minimum completion time - assuming that we have an unlimited number of workers, what is the fastest we can get this job completed while respecting precedence constraints. If there were no precedence constraints, each task could be worked on by its own worker, and the total time would be that of the longest single task. If there were such strict precedence constraints that each task had to follow the completion of its immediate predecessor, the minimum completion time would be obtained by summing up the times for each task. The minimum completion time for a DAG can be easily computed in O(n+m) time. Initialize the completion time for all the vertices to 0, except the start vertex, which is initialized to the length of the start task. Perform a topological sort to order the vertices such that all precedences will have been considered by the time we evaluate each vertex. For each vertex u, consider all the edges (u,v) leaving u. The completion time of these vertices is the maximum of the current completion time for v plus the completion time of u plus the task time of v. Critical path - The longest path from the start vertex to the completion vertex defines the critical path. This can be important to know, for the only way to shorten the minimum total completion time is to reduce the time of one of the tasks on each critical path. The tasks on the critical paths can be determined in O(n+m) time using the simple dynamic programming presented in Section . What is the tradeoff between number of workers and completion time? - What we would really be interested in knowing is how best to complete the schedule with a given number of workers. Unfortunately, this and most similar problems are NP-complete.

file:///E|/BOOK/BOOK4/NODE156.HTM (2 of 4) [19/1/2003 1:30:50]

Job Scheduling

An even more general formulation seeks critical paths in networks where certain jobs are restricted to certain people. Such networks are known as disjunctive networks. Finding critical paths on such networks is more complicated than CPM/PERT, but implementations are described below. In any sufficiently large and sufficiently real scheduling application, there will be combinations of constraints that are difficult or impossible to model using these techniques. There are two reasonable ways to deal with such problems. First, we can ignore enough constraints that the problem reduces to one of the types that we have described here, solve it, and then see how bad it is using the other constraints. Perhaps the schedule can be easily modified by hand to satisfy constraints like keeping Joe and Bob apart so they can't kill each other. Another approach is to formulate your scheduling problem via linear-integer programming (see Section ) in all its complexity. This method can be better only if you really know enough about your desired schedule to formulate the right linear program, and if you have the time to wait for the program to give you the solution. I would start out with something simpler and see what happens first. Another fundamental scheduling problem takes a set of jobs without precedence constraints and assign them to identical machines so as to minimize the total elapsed time. Consider a copy shop with k Xerox machines and a stack of jobs to finish by the end of the day. Such tasks are called job-shop scheduling. ), where each job is assigned a number They can be modeled as bin-packing problems (see Section equal to the number of hours it will take and each machine is represented by a bin with space equal to the number of hours in a day. More sophisticated variations of job-shop scheduling provide each task with allowable start and required finishing times. Effective heuristics are known, based on sorting the tasks by size and finishing time. We refer the reader to the references for more information. Note that these scheduling problems become hard only when the tasks cannot be broken up onto multiple machines or interrupted (preempted) and then rescheduled. If your application has these degrees of freedom, you should exploit them. Implementations: Pascal implementations of Balas's algorithm for disjunctive network scheduling and Hu's algorithm for assigning jobs to processors with precedence constraints appear in [SDK83]. See Section

.

Algorithm 520 [WBCS77] of the Collected Algorithms of the ACM is a Fortran code for multipleresource network scheduling. It is available from Netlib (see Section

).

Notes: The literature on scheduling algorithms is a vast one. For a more detailed survey of the field, we refer the reader to [Cof76, LLK83]. Good expositions on CPM/PERT include [Eve79a, Law76, PGD82, SDK83]. Good expositions on jobfile:///E|/BOOK/BOOK4/NODE156.HTM (3 of 4) [19/1/2003 1:30:50]

Job Scheduling

shop scheduling include [AC91, SDK83]. Related Problems: Topological sorting (see page page

), edge coloring (see page

), matching (see page

), bin packing (see page

), vertex coloring (see

).

Next: Satisfiability Up: Combinatorial Problems Previous: Calendrical Calculations Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE156.HTM (4 of 4) [19/1/2003 1:30:50]

Satisfiability

Next: Graph Problems: Polynomial-Time Up: Combinatorial Problems Previous: Job Scheduling

Satisfiability

Input description: A set of clauses in conjunctive normal form. Problem description: Is there a truth assignment to the Boolean variables such that every clause is simultaneously satisfied? Discussion: Satisfiability arises whenever we seek a configuration or object that must be consistent with (i.e. satisfy) a given set of constraints. For example, consider the problem of drawing name labels for cities on a map. For the labels to be legible, we do not want the labels to overlap, but in a densely populated region many labels need to be drawn in a small space. How can we avoid collisions? For each of the n cities, suppose we identify two possible places to position its label, say right above or right below each city. We can represent this choice by a Boolean variable , which will be true if city . Certain pairs of labels may be forbidden, such as when 's 's label is above , otherwise above label would obscure 's below label. This pairing can be forbidden by the two-element clause , where means ``not v''. Finding a satisfying truth assignment for the resulting set of clauses yields a mutually legible map labeling if one exists. Satisfiability is the original NP-complete problem. Despite its applications to constraint satisfaction, file:///E|/BOOK/BOOK4/NODE157.HTM (1 of 4) [19/1/2003 1:30:52]

Satisfiability

logic, and automatic theorem proving, it is perhaps most important theoretically as the root problem from which all other NP-completeness proofs originate. ●

Is your formula in CNF or DNF? - In satisfiability, the constraints are specified as a logical formula. There are two primary ways of expressing logical formulas, conjunctive normal form (CNF) and disjunctive normal form (DNF). In CNF formulas, we must satisfy all clauses, where each clause is constructed by and-ing or's of literals together, such as

In DNF formulas, we must satisfy any one clause, where each clause is constructed by or-ing ands of literals together. The formula above can be written in DNF as

●

●

Solving DNF-satisfiability is trivial, since any DNF formula can be satisfied unless every clause contains both a literal and its complement (negation). However, CNF-satisfiability is NPcomplete. This seems paradoxical, since we can use De Morgan's laws to convert CNF-formulae into equivalent DNF-formulae and vice versa. The catch is that an exponential number of terms might be constructed in the course of translation, so that the translation itself might not run in polynomial time. How big are your clauses? - k-SAT is a special case of satisfiability when each clause contains at most k literals. The problem of 1-SAT is trivial, since we must set true any literal appearing in any clause. The problem of 2-SAT is not trivial, but it can still be solved in linear time. This is very interesting, because many problems can be modeled as 2-SAT using a little cleverness. Observe that the map labeling problem described above is an instance of 2-SAT and hence can be solved in time linear in the number of clauses, which might be quadratic in the number of variables. The good times end as soon as clauses contain three literals each, i.e. 3-SAT, for 3-SAT is NPcomplete. Thus in general it will not be helpful to model a problem as satisfiability unless we can do it with two-element clauses. Does it suffice to satisfy most of the clauses? - Given an instance of general satisfiability, there is not much you can do to solve it except by backtracking algorithms such as the Davis-Putnam procedure. In the worst case, there are truth assignments to be tested, but fortunately, there are lots of ways to prune the search. Although satisfiability is NP-complete, how hard it is in practice depends on how the instances are generated. Naturally defined ``random'' instances are often surprisingly easy to solve, and in fact it is nontrivial to generate instances of the problem that are truly hard. Still, we would likely benefit by relaxing the problem so that the goal is to satisfy as many clauses as possible. Here optimization techniques such as simulated annealing can be put to work to refine random or heuristic solutions. Indeed, any random truth assignment to the variables will satisfy

file:///E|/BOOK/BOOK4/NODE157.HTM (2 of 4) [19/1/2003 1:30:52]

Satisfiability

any particular k-SAT clause with probability , so our first attempt is likely to satisfy most of the clauses. Finishing off the job is the hard part. Finding an assignment that satisfies the maximum number of clauses is NP-complete even for nonsatisfiable instances. When faced with a problem of unknown complexity, proving the problem NP-complete can be an important first step. If you think your problem might be hard, the first thing to do is skim through Garey and Johnson [GJ79] looking for your problem. If you don't find it, my recommendation is to put the book away and try to prove hardness from first principles, using one of the basic problems in this catalog, particularly 3-SAT, vertex cover, independent set, integer partition, clique, and Hamiltonian cycle. I find it much easier to start from these than some complicated problem I stumble over in the book, and more insightful too, since the reason for the hardness is not obscured by the hidden hardness proof for the complicated problem. Chapter

focuses on strategies for proving hardness.

Implementations: Programs for solving satisfiability problems were sought for the Second DIMACS Implementation Challenge, held in October 1993. Programs and data from the challenge are available by anonymous ftp from dimacs.rutgers.edu in the directory /pub/challenge/sat. In particular, sato is a decision procedure for propositional logic written in C by Hantao Zhang. There is also a random formula generator named mwff.c for constructing hard satisfiability instances in C by Bart Selman. Several other solvers and instance generators are also available from this site. The propositional satisfiability tester POSIT, by Jon W. Freeman, is based on a highly optimized version of the Davis-Putnum procedure. It is available by anonymous ftp from ftp.cis.upenn.edu in /pub/freeman/posit-1.0.tar.Z. Notes: The primary reference on NP-completeness is [GJ79], featuring a list of roughly four hundred NPcomplete problems. Although the list is now fifteen years old, it remains an extremely useful reference; it is perhaps the book I reach for most often. An occasional column by David Johnson in the Journal of Algorithms has helped to update the book. In [Joh90], Johnson gives a thorough and readable survey of the relationship between different complexity classes. Good expositions of Cook's theorem [Coo71], where satisfiability is proven hard, include [CLR90, GJ79, PS82]. The importance of Cook's result became clear in Karp's paper [Kar72] showing that it implied the hardness of over twenty different combinatorial problems. A linear-time algorithm for 2-satisfiability appears in [APT79]. The application of 2-satisfiability to map labeling is taken from [WW95]. Related Problems: Constrained optimization (see page

file:///E|/BOOK/BOOK4/NODE157.HTM (3 of 4) [19/1/2003 1:30:52]

), traveling salesman problem (see page

).

Satisfiability

Next: Graph Problems: Polynomial-Time Up: Combinatorial Problems Previous: Job Scheduling Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE157.HTM (4 of 4) [19/1/2003 1:30:52]

Graph Problems: Polynomial-Time

Next: Connected Components Up: A Catalog of Algorithmic Previous: Satisfiability

Graph Problems: Polynomial-Time Algorithmic graph problems constitute approximately one third of all the problems in this catalog. Further, several problems from other sections can be formulated strictly in terms of graphs. Identifying the name of a graph-theoretic invariant or problem is one of the primary skills of a good algorist. Indeed, this catalog will tell you exactly how to proceed as soon as you figure out your particular problem's name. We have partitioned the bulk of the algorithmic graph problems in this book between this and the subsequent section. Here, we deal only with problems for which there exist efficient algorithms to solve them. As there is often more than one way to model a given application, it makes sense to look here before proceeding on to the harder formulations. The algorithms presented here have running times that grow slowly with the size of the graph. We adopt throughout the standard convention that n refers to the number of vertices in a graph, while m is the number of edges. Although graphs are combinatorial objects, describing a binary relation on a set of objects, graphs are usually best understood as drawings. Beyond just the problems of visualization, many interesting graph properties follow from the nature of a particular type of drawing, such as planar graphs. In this chapter, we also present a variety of different problems and algorithms associated with graph drawing. Most advanced graph algorithms are difficult to program. However, good implementations are often available if you know where to look. The best single source is LEDA, discussed in Section faster special-purpose codes exist for many problems.

, although

Books specializing in graph algorithms include: ●

●

Even [Eve79a] - This is a good book on graph algorithms, fairly advanced, with a particularly thorough treatment of planarity-testing algorithms. Ahuja, Magnanti, and Orlin [AMO93] - While purporting to be a book on network flows, it covers the gamut of graph algorithms with emphasis on operations research. Strongly

file:///E|/BOOK/BOOK4/NODE158.HTM (1 of 2) [19/1/2003 1:30:53]

Graph Problems: Polynomial-Time

●

●

● ● ● ● ● ● ● ● ● ● ● ●

recommended. van Leeuwen [vL90a] - A 100+ page survey on research results in graph algorithms, this is the best source to determine what is known in algorithmic graph theory. McHugh [McH90] - A more elementary but comprehensive treatment of basic graph algorithms, including parallel algorithms.

Connected Components Topological Sorting Minimum Spanning Tree Shortest Path Transitive Closure and Reduction Matching Eulerian Cycle / Chinese Postman Edge and Vertex Connectivity Network Flow Drawing Graphs Nicely Drawing Trees Planarity Detection and Embedding

Next: Connected Components Up: A Catalog of Algorithmic Previous: Satisfiability Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE158.HTM (2 of 2) [19/1/2003 1:30:53]

Connected Components

Next: Topological Sorting Up: Graph Problems: Polynomial-Time Previous: Graph Problems: Polynomial-Time

Connected Components

Input description: A directed or undirected graph G. Problem description: Traverse each edge and vertex of all connected components of G. Discussion: The connected components of a graph represent, in grossest terms, the pieces of the graph. Two vertices are in the same component of G if and only if there is some path between them. Finding connected components is at the heart of many graph applications. For example, consider the problem of identifying clusters in a set of items. We can represent each item by a vertex and add an edge between each pair of items that are deemed ``similar.'' The connected components of this graph correspond to different classes of items. Testing whether a graph is connected is an essential preprocessing step for every graph algorithm. Such tests can be performed so quickly and easily that you should always verify that your input graph is connected, even when you know it has to be. Subtle, difficult-to-detect bugs often result when your algorithm is run only on one component of a disconnected graph.

file:///E|/BOOK/BOOK4/NODE159.HTM (1 of 4) [19/1/2003 1:30:54]

Connected Components

Testing the connectivity of any undirected graph is a job for either depth-first or breadth-first search, as . Which one you choose doesn't really matter. Both traversals initialize a discussed in Section component-number field for each vertex to 0, and then start the search for component 1 from vertex . As each vertex is visited, the value of this field is set to the current component number. When the initial traversal ends, the component number is incremented, and the search begins again from the first vertex with component-number still 0. Properly implemented using adjacency lists, this runs in O(n+m), or time linear in the number of edges and vertices. Other notions of connectivity also arise in practice: ●

What if my graph is directed? - There are two distinct definitions of connected components for directed graphs. A directed graph is weakly connected if it would be connected by ignoring the direction of edges. Thus a weakly connected graph consists of a single piece. A directed graph is strongly connected if there is a directed path between every pair of vertices. This distinction is best made clear by considering a network of one- and two-way streets in a city. The network is strongly connected if it is possible to drive legally between every two positions. The network is weakly connected if it is possible to drive legally or illegally between every two positions. The network is disconnected if there is no possible way to drive from a to b. The weakly and strongly connected components define unique partitions on the vertices. The output figure above illustrates a directed graph consisting of two weakly connected or five strongly connected components (also called blocks of G). Testing whether a directed graph is weakly connected can be done easily in linear time. Simply turn all edges into undirected edges and use the DFS-based connected components algorithm described above. Tests for strong connectivity are somewhat more complicated. The simplest algorithm performs a breadth-first search from each vertex and verifies that all vertices have been visited on each search. Thus in O(mn) time, it can be confirmed whether the graph is strongly connected. Further, this algorithm can be modified to extract all strongly connected components if it is not.

●

In fact, strongly connected components can be found in linear time using one of two more sophisticated DFS-based algorithms. See the references below for details. It is probably easier to start from an existing implementation below than a textbook description. How reliable is my network; i.e. how well connected is it? - A chain is only as strong as its weakest link. When it is missing one or more links, it is disconnected. The notion of connectivity of graphs measures the strength of the graph - how many edges or vertices must be removed in order to break it, or disconnect it. Connectivity is an essential invariant for network design and other structural problems. . In particular, biconnected Algorithmic connectivity problems are discussed in Section components are pieces of the graph that result by cutting the edges incident on a single vertex. All

file:///E|/BOOK/BOOK4/NODE159.HTM (2 of 4) [19/1/2003 1:30:54]

Connected Components

●

biconnected components can be found in linear time using a DFS-based algorithm. Vertices whose deletion disconnects the graph belong to more than one biconnected component, although edges are uniquely partitioned among them. Is the graph a tree? How can I find a cycle if one exists? - The problem of cycle identification often arises, particularly with respect to directed graphs. For example, testing if a sequence of conditions can deadlock often reduces to cycle detection. If I am waiting for Fred, and Fred is waiting for Mary, and Mary is waiting for me, we are all deadlocked. For undirected graphs, the analogous problem is tree identification. A tree is, by definition, an undirected, connected graph without any cycles. As described above, a depth-first search can be used to test whether it is connected. If the graph is connected and has n-1 edges for n vertices, it is a tree. Depth-first search can be used to find cycles in both directed and undirected graphs. Whenever we encounter a back edge in our DFS, i.e. an edge to an ancestor vertex in the DFS tree, the back edge and the tree together define a directed cycle. No other such cycle can exist in a directed graph. Directed graphs without cycles are called DAGs (directed acyclic graphs). Topological sorting (see Section

) is the fundamental operation on DAGs.

Implementations: LEDA (see Section ) provides good implementations of breadth-first and depthfirst search, connected components and strongly connected components, all in C++. XTango (see Section ) is an algorithm animation system for UNIX and X-windows, which includes an animation of depthfirst search. Pascal implementations of BFS, DFS, and biconnected and strongly connected components appear in [MS91]. See Section

for details. Combinatorica [Ski90] provides Mathematica implementations of

the same routines. See Section

.

) contains routines to compute biconnected and strongly The Stanford GraphBase (see Section connected components. An expository implementation of BFS and DFS in Fortran appears in [NW78] (see Section

).

Notes: Depth-first search was first used in algorithms for finding paths out of mazes, and dates back to the nineteenth century [Luc91, Tar95]. Breadth-first search was first reported to find the shortest path out of mazes by Moore in 1957 [Moo59]. Hopcroft and Tarjan [HT73b, Tar72] first established depth-first search as a fundamental technique for efficient graph algorithms. Expositions on depth-first and breadth-first search appear in every book

file:///E|/BOOK/BOOK4/NODE159.HTM (3 of 4) [19/1/2003 1:30:54]

Connected Components

discussing graph algorithms, with [CLR90] perhaps the most thorough description available. The first linear-time algorithm for strongly connected components is due to Tarjan [Tar72], with expositions including [Baa88, Eve79a, Man89]. Another algorithm, simpler to program and slicker, to find strongly connected components is due to Sharir and Kosaraju. Good expositions of this algorithm appear in [AHU83, CLR90]. Related Problems: Edge-vertex connectivity (see page

), shortest path (see page

).

Next: Topological Sorting Up: Graph Problems: Polynomial-Time Previous: Graph Problems: Polynomial-Time Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE159.HTM (4 of 4) [19/1/2003 1:30:54]

Topological Sorting

Next: Minimum Spanning Tree Up: Graph Problems: Polynomial-Time Previous: Connected Components

Topological Sorting

Input description: A directed acyclic graph G=(V,E), also known as a partial order or poset. Problem description: Find a linear ordering of the vertices of V such that for each edge i is to the left of vertex j.

, vertex

Discussion: Topological sorting arises as a natural subproblem in most algorithms on directed acyclic graphs. Topological sorting orders the vertices and edges of a DAG in a simple and consistent way and hence plays the same role for DAGs that depth-first search does for general graphs. Topological sorting can be used to schedule tasks under precedence constraints. Suppose we have a set of tasks to do, but certain tasks have to be performed before other tasks. These precedence constraints form a directed acyclic graph, and any topological sort (also known as a linear extension) defines an order to do these tasks such that each is performed only after all of its constraints are satisfied. Three important facts about topological sorting are: ●

Only directed acyclic graphs can have linear extensions, since any directed cycle is an inherent

file:///E|/BOOK/BOOK4/NODE160.HTM (1 of 3) [19/1/2003 1:30:55]

Topological Sorting

●

●

contradiction to a linear order of tasks. Every DAG can be topologically sorted, so there must always be at least one schedule for any reasonable precedence constraints among jobs. DAGs typically allow many such schedules, especially when there are few constraints. Consider n jobs without any constraints. Any of the n! permutations of the jobs constitutes a valid linear extension.

A linear extension of a given DAG is easily found in linear time. The basic algorithm performs a depthfirst search of the DAG to identify the complete set of source vertices, where source vertices are vertices without incoming edges. At least one such source must exist in any DAG. Note that source vertices can appear at the start of any schedule without violating any constraints. After deleting all the outgoing edges of the source vertices, we will create new source vertices, which can sit comfortably to the immediate right of the first set. We repeat until all vertices have been accounted for. Only a modest amount of care with data structures (adjacency lists and queues) is needed to make this run in O(n+m) time. This algorithm is simple enough that you should be able to code up your own implementation and expect good performance, although implementations are described below. Two special considerations are: ●

●

What if I need all the linear extensions, instead of just one of them? - In certain applications, it is important to construct all linear extensions of a DAG. Beware, because the number of linear extensions can grow exponentially in the size of the graph. Even the problem of counting the number of linear extensions is NP-hard. Algorithms for listing all linear extensions in a DAG are based on backtracking. They build all possible orderings from left to right, where each of the in-degree zero vertices are candidates for the next vertex. The outgoing edges from the selected vertex are deleted before moving on. Constructing truly random linear extensions is a hard problem, but pseudorandom orders can be constructed from left to right by selecting randomly among the in-degree zero vertices. What if your graph is not acyclic? - When the set of constraints is not a DAG, but it contains some inherent contradictions in the form of cycles, the natural problem becomes to find the smallest set of jobs or constraints that if eliminated leaves a DAG. These smallest sets of offending jobs (vertices) or constraints (edges) are known as the feedback vertex set and the feedback arc set, respectively, and are discussed in Section NP-complete problems.

. Unfortunately, both of them are

Since the basic topological sorting algorithm will get stuck as soon as it identifies a vertex on a directed cycle, we can delete the offending edge or vertex and continue. This quick-and-dirty heuristic will eventually leave a DAG. Implementations: Many textbooks contain implementations of topological sorting, including [MS91]

file:///E|/BOOK/BOOK4/NODE160.HTM (2 of 3) [19/1/2003 1:30:55]

Topological Sorting

(see Section ) and [Sed92] (see Section ). implementation of topological sorting in C++.

LEDA (see Section

) includes a linear-time

) is an algorithm animation system for UNIX and X-windows, which includes XTango (see Section an animation of topological sorting. Combinatorica [Ski90] provides Mathematica implementations operations on directed acyclic graphs. See Section

of topological sorting and other

.

Notes: Good expositions on topological sorting include [CLR90, Man89]. Brightwell and Winkler [BW91] proved that it is #P-complete to count the number of linear extensions of a partial order. The complexity class #P includes NP, so any #P-complete problem is at least NP-hard. Related Problems: Sorting (see page

), feedback edge/vertex set (see page

).

Next: Minimum Spanning Tree Up: Graph Problems: Polynomial-Time Previous: Connected Components Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE160.HTM (3 of 3) [19/1/2003 1:30:55]

Minimum Spanning Tree

Next: Shortest Path Up: Graph Problems: Polynomial-Time Previous: Topological Sorting

Minimum Spanning Tree

Input description: A graph G = (V,E) with weighted edges. Problem description: The subset of

of minimum weight forming a tree on V.

Discussion: The minimum spanning tree (MST) of a graph defines the cheapest subset of edges that keeps the graph in one connected component. Telephone companies are particularly interested in minimum spanning trees, because the minimum spanning tree of a set of sites defines the wiring scheme that connects the sites using as little wire as possible. It is the mother of all network design problems. Minimum spanning trees prove important for several reasons: ●

●

● ●

They can be computed quickly and easily, and they create a sparse subgraph that reflects a lot about the original graph. They provide a way to identify clusters in sets of points. Deleting the long edges from a minimum spanning tree leaves connected components that define natural clusters in the data set, as shown in the output figure above. They can be used to give approximate solutions to hard problems such as Steiner tree and traveling salesman. As an educational tool, minimum spanning tree algorithms provide graphic evidence that greedy algorithms can give provably optimal solutions.

Two classical algorithms efficiently construct minimum spanning trees, namely Prim's and Kruskal's. Brief overviews of both algorithms are given below, with correctness arguments in Section implementation details.

. We refer the reader to the codes for

Prim's algorithm starts with an arbitrary vertex v and ``grows'' a tree from it, repeatedly finding the lowest-cost edge that will link some new vertex into this tree. During execution we will label each vertex as either in the tree, fringe - meaning there exists an edge from a tree vertex, or unseen - meaning the vertex is more than one edge away from the tree.

file:///E|/BOOK/BOOK4/NODE161.HTM (1 of 4) [19/1/2003 1:30:57]

Minimum Spanning Tree

Prim(G) Select an arbitrary vertex to start

While (there are fringe vertices)

select minimum-weight edge between tree and fringe

add the selected edge and vertex to the tree

This creates a spanning tree for any connected graph, since no cycle can be introduced via edges between tree and fringe vertices. That it is in fact a tree of minimum weight can be proven by contradiction, and the proof is in Section time. simple data structures, Prim's algorithm can be implemented in

. With

Kruskal's algorithm is also greedy. It starts with each vertex as a separate tree and merges these trees together by repeatedly adding the lowest cost edge that merges two distinct subtrees (i.e. does not create a cycle).

Kruskal(G) Sort the edges in order of increasing weight

count=0

while (count < n-1) do

get next edge (v,w)

if (component (v)

component(w))

add to T

component(v) = component(w)

The ``which component?'' tests are efficiently implemented using the union-find data structure of Section algorithm.

, to yield an

Minimum spanning tree is only one of several spanning tree problems that arise in practice. The following questions will file:///E|/BOOK/BOOK4/NODE161.HTM (2 of 4) [19/1/2003 1:30:57]

Minimum Spanning Tree

help you sort your way through them: ●

●

●

●

Are the weights of all edges of your graph identical? - Every spanning tree on n points contains exactly n-1 edges. Thus if your graph is unweighted, any spanning tree will be a minimum spanning tree. Either breadth-first or depth-first search can be used to find a rooted spanning tree in linear time. Depth-first search trees tend to be long . and thin, while breadth-first search trees better reflect the distance structure of the graph, as discussed in Section Should I use Prim's or Kruskal's algorithm? - As described, Prim's algorithm runs in , while Kruskal's time. Thus Prim's algorithm is faster on dense graphs, while Kruskal's is faster on sparse algorithm takes graphs. Although Prim's algorithm can be implemented in time using more advanced data structures (in particular, Fibonacci heaps), this will not be worth the trouble unless you have extremely large, fairly sparse graphs. I personally find Kruskal's algorithm easier to understand and implement than Prim's, but that is just a matter of taste. What if my input is points in the plane, instead of a graph? - Geometric instances, comprising n points in dand then finding the MST of this dimensions, can be solved by constructing the complete distance graph in complete graph. However, for points in two or even three dimensions, it can be more efficient to solve the geometric version of the problem directly. To find the minimum spanning tree of n points, first construct the Delaunay and ). In two dimensions, this gives a graph with O(n) edges that triangulation of these points (see Sections contains all the edges of the minimum spanning tree of the point set. Running Kruskal's algorithm on this sparse time. graphs finishes the job in How do I find a spanning tree that avoids vertices of high degree? - Another common goal of spanning tree problems is to minimize the maximum degree, typically to minimize the fan out in an interconnection network. Unfortunately, finding a spanning tree of maximum degree 2 is clearly NP-complete, since this is identical to the Hamiltonian path problem. Efficient algorithms are known, however, that construct spanning trees whose maximum degree at most one more than required. This is likely to suffice in practice. See the references below.

Implementations: Pascal implementations of Prim's, Kruskal's, and the Cheriton-Tarjan algorithm are provided in [MS91], along with extensive empirical analysis that shows that the implementation of Prim's algorithm with the appropriate priority queue is fastest on most graphs. See Section

.

) contains implementations of four different minimum spanning tree algorithms, The Stanford GraphBase (see Section and the result of timing experiments suggesting that Kruskal's algorithm is best. The results are reported in terms of memory accesses (mems) instead of seconds, to make them independent of processor speed. A C++ implementation of Kruskal's algorithm is provided in LEDA (see Section

). Alternative implementations of

Prim's and Kruskal's algorithms are provided in Pascal [SDK83] and C++ [Sed92]. See Section Section Section

. XTango (see

) includes an animation of both Prim's and Kruskal's algorithms.

Algorithm 479 [Pag74] and Algorithm 613 [HJS84] of the Collected Algorithms of the ACM are Fortran codes for minimum spanning tree, the former in an implementation of a point clustering algorithm. They are available from Netlib (see Section

). A bare bones Fortran implementation is provided in [NW78], including the enumeration of all spanning trees.

See Section

.

Combinatorica [Ski90] provides Mathematica implementations of Kruskal's minimum spanning tree algorithm and quickly

file:///E|/BOOK/BOOK4/NODE161.HTM (3 of 4) [19/1/2003 1:30:57]

Minimum Spanning Tree

counting the number of spanning trees of a graph.

See Section

.

Notes: Good expositions on Prim's [Pri57] and Kruskal's [Kru56] algorithms will appear in any textbook on algorithms, but include [Baa88, CLR90, Man89, Tar83]. The fastest implementations of Prim's and Kruskal's algorithms use Fibonacci heaps [FT87]. Expositions of faster algorithms for geometric instances include [PS85]. A recent breakthrough on the minimum spanning tree problem is the linear-time randomized algorithm of Karger, Klein, and Tarjan [KKT95]. Simplifications will be needed before this becomes the algorithm of choice. The history of the minimum spanning tree problem dates back at least to Boruvka, in 1926, and is presented in [GH85]. Interestingly, it is Boruvka's algorithm that serves as the foundation to the new randomized one. Fürer and Raghavachari [FR94] give an algorithm that constructs a spanning tree whose maximum degree is almost minimized, indeed is at most one more than the lowest-degree spanning tree. The situation is analogous to Vizing's theorem for edge coloring, which also gives an approximation algorithm to within additive factor one. Minimum spanning tree algorithms have an interpretation in terms of matroids, which are systems of subsets closed under inclusion, for which the maximum weighted independent set can be found using a greedy algorithm. The connection between greedy algorithms and matroids was established by Edmonds [Edm71]. Expositions on the theory of matroids include [Law76, PS82]. Algorithms for generating spanning trees in order from minimum to maximum weight are presented in [Gab77]. Good expositions on the matrix-tree theorem, which counts the number of spanning trees of a graph, include [Eve79a]. Related Problems: Steiner tree (see page

), traveling salesman (see page

).

Next: Shortest Path Up: Graph Problems: Polynomial-Time Previous: Topological Sorting Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE161.HTM (4 of 4) [19/1/2003 1:30:57]

Shortest Path

Next: Transitive Closure and Reduction Up: Graph Problems: Polynomial-Time Previous: Minimum Spanning Tree

Shortest Path

Input description: An edge-weighted graph G, with start vertex and end vertex t. Problem description: Find the shortest path from to t in G. Discussion: The problem of finding shortest paths in a graph has a surprising variety of applications: ●

●

●

●

The most obvious applications arise in transportation or communications, such as finding the best route to drive between Chicago and Phoenix or figuring how to direct packets to a destination across a network. Consider the problem of image segmentation, that is, separating two characters in a scanned, bit-mapped image of printed text. We need to find the separating line between two points that cuts through the fewest number of black pixels. This grid of pixels can be modeled as a graph, with any edge across a black pixel given a high cost. The shortest path from top to bottom defines the best separation between left and right. A major problem in speech recognition is distinguishing between words that sound alike (homophones), such as to, two, and too. We can construct a graph whose vertices correspond to possible words, with an edge between possible neighboring words. If the weight of each edge measures the likelihood of transition, the shortest path across the graph defines the best interpretation of a sentence. For a more detailed account of such an application, . see Section Suppose we want to draw an informative picture of a graph. The center of the page should coorespond to the ``center'' of the graph, whatever that means. A good definition of the center is the vertex that minimizes the maximum distance to any other vertex in the graph. Finding this center point requires knowing the distance (i.e. shortest path) between all pairs of vertices.

The primary algorithm for finding shortest paths is Dijkstra's algorithm, which efficiently finds the shortest paths from a given vertex x to all n-1 other vertices. Dijkstra's algorithm starts from x. In each iteration, it identifies a new vertex v for which the shortest path from x to v is known. We maintain a set of vertices S to which we currently know the shortest path from v, and this set grows by one vertex in each iteration. In each iteration, we identify the edge (u,v) where

file:///E|/BOOK/BOOK4/NODE162.HTM (1 of 5) [19/1/2003 1:31:00]

Shortest Path

and

such that

This edge (u,v) gets added to a shortest path tree, whose root is x and which describes all the shortest paths from x. See the discussion in Section

for more details.

The straightforward implementation of this algorithm is O(m n). However, with simple data structures it can be reduced to or time. Theoretically faster times can be achieved using significantly more complicated data structures, as described below. If we are just interested in knowing the shortest path from x to y, simply stop as soon as y enters S. Dijkstra's algorithm is the right choice for single-source shortest path on positively weighted graphs. However, special circumstances dictate different choices: ●

●

●

●

●

Is your graph weighted or unweighted? - If your graph is unweighted, a simple breadth-first search starting from the source vertex will find the shortest path in linear time. It is only when edges have different weights that you need more sophisticated algorithms. Breadth-first search is both simpler and faster than Dijkstra's algorithm. Does your graph have negative cost weights? - Dijkstra's algorithm assumes that all edges have positive cost. If your graph has edges with negative weights, you must use the more general but less efficient Bellman-Ford algorithm. If your graph has a cycle of negative cost, than the shortest path between any two vertices in the graph is not defined, since we can detour to the negative cost cycle and loop around it an arbitrary number of times, making the total cost as small as desired. Note that adding the same amount of weight to each edge to make it positive and running Dijkstra's algorithm does not find the shortest path in the original graph, since paths that use more edges will be rejected in favor of longer paths using fewer edges. Why might one ever need to find shortest paths in graphs with negative cost edges? An interesting application comes in currency speculation. Construct a graph where each vertex is a nation and there is an edge weighted from x to y if the exchange rate from currency x to currency y is w(x,y). In arbitrage, we seek a cycle to convert currencies so that we end up with more money than we started with. For example, if the exchange rates are 12 pesos per dollar, 5 pesos per franc, and 2 francs per dollar, by simply moving money around we can convert $1 to $1.20. In fact, there will be a profit opportunity whenever there exists a negative cost cycle in this weighted graph. Is your input a set of geometric obstacles instead of a graph? - If you seek the shortest path between two points in a geometric setting, like an obstacle-filled room, you may either convert your problem into a graph of distances and feed it to Dijkstra's algorithm or use a more efficient geometric algorithm to compute the shortest path directly on motion planning. from the arrangement of obstacles. For such geometric algorithms, see Section Does your graph have cycles in it, or is it a DAG? - If your graph is a directed acyclic graph (DAG), than the shortest path can be found in linear time. Perform a topological sort to order the vertices such that all edges go from left to right, then do dynamic programming on the left-to-right ordered vertices. Indeed, most dynamic programming problems can be easily formulated as shortest paths on specific DAGs. The algorithm is discussed in [Man89] if you cannot figure it out from this description. Note that the same algorithm (replacing with ) also suffices for finding the longest path in a DAG. which is useful in many applications like scheduling (see ). Section Do you need the shortest path between all pairs of points? - If you are interested in the shortest path between all pairs of vertices, one solution is to run Dijkstra n times, once with each vertex as the source. However, the FloydWarshall algorithm is a slick dynamic programming algorithm for all-pairs shortest path, which is faster and

file:///E|/BOOK/BOOK4/NODE162.HTM (2 of 5) [19/1/2003 1:31:00]

Shortest Path

easier to program than Dijkstra and which works with negative cost edges (but not cycles). It is discussed more . Let M denote the distance matrix, where

thoroughly in Section

if there is no edge (i,j).

for k = 1 to n do

for i = 1 to n do

for j = 1 to n do

Return

●

denotes ``the length of the shortest path from i to j that goes The key to understanding Floyd's algorithm is that through at most k intermediate vertices.'' Note that space suffices, since we need keep only and around at time k. How do I find the shortest cycle in a graph? One application of all-pairs shortest path is to find the shortest , which is the length of cycle in a graph, called its girth. Floyd's algorithm can be used to compute for the shortest way to get from vertex i to i, in other words the shortest cycle through i. This might be exactly what you want. However, the shortest cycle through x is likely to go from x to y back to x, using the same edge twice. A simple cycle is one that visits no edge or vertex twice. To find the shortest simple cycle, the easiest approach is to compute the lengths of the shortest paths from i to all other vertices, and then explicitly check whether there is an acceptable edge from each vertex back to i.

Figure: The girth, or shortest cycle, in a graph Finding the longest cycle in a graph includes the special case of Hamiltonian cycle (see

file:///E|/BOOK/BOOK4/NODE162.HTM (3 of 5) [19/1/2003 1:31:00]

), so it is NP-

Shortest Path

complete. The all-pairs shortest path matrix can be used to compute several useful invariants of any graph G, that are related to the center of G. The eccentricity of a vertex v in a graph is the shortest-path distance to the farthest vertex from v. From the eccentricity come other graph invariants. The radius of a graph is the smallest eccentricity of any vertex, while the center is the set of vertices whose eccentricity is the radius. The diameter of a graph is the maximum eccentricity of any vertex.

Implementations: The highest performance code (for both Dijkstra and Bellman-Ford) available for finding shortest paths in graphs is SPLIB [CGR93], developed in C language by Cherkassky, Goldberg, and Radzik. They report solving instances with over one million vertices in under two minutes on a Sun Sparc-10 workstation. Their codes are available from http://www.neci.nj.nec.com/homepages/avg.html for noncommercial use. ) provides good implementations in C++ for all of the shortest-path algorithms we have LEDA (see Section discussed, including Dijkstra, Bellman-Ford, and Floyd's algorithms. Pascal implementations of Dijkstra, Bellman-Ford, and Floyd's algorithms are given in [SDK83]. See Section

.

) includes animations of both Dijkstra's and Floyd's shortest-path algorithms.

XTango (see Section

Combinatorica [Ski90] provides Mathematica implementations

of Dijkstra's and Floyd's algorithms for shortest paths,

acyclicity testing, and girth computation for directed/undirected and weighted/unweighted graphs. See Section

.

The Stanford GraphBase (see Section ) contains an implementation of Dijkstra's algorithm, used for computing word ladders in a graph defined by five-letter words, as well as an implementation of a program to bound the girth of graphs. Algorithm 562 [Pap80] of the Collected Algorithms of the ACM is a Fortran program to find shortest paths in graphs (see Section

).

Notes: Good expositions on Dijkstra's algorithm [Dij59] and Floyd's all-pairs-shortest-path algorithm [Flo62] include [Baa88, CLR90, Man89]. Good expositions of the Bellman-Ford algorithm [Bel58, FF62] are slightly rarer, but include [CLR90, Eve79a, Law76]. Expositions on finding the shortest path in a DAG include [Law76]. A survey on shortest-path algorithms with 222 references appears in [DP84]. Included are citations to algorithms for related path problems, like finding the kth-shortest path and shortest paths when edge costs vary with time. Expositions on finding the kth-shortest path include [Law76]. The theoretically fastest algorithms known for single-source shortest path for positive edge weight graphs are variations of Dijkstra's algorithm with Fibonacci heaps [FT87]. Experimental studies of shortest-path algorithms include [DF79, DGKK79]. However, these experiments were done before Fibonacci heaps were developed. See [CGR93] for a more recent study. Theoretically faster algorithms exist when the weights of the edges are small; i.e. their absolute values are each bounded by W. For positive edge weights, the single-source-shortest-path can be found in [AMOT88], while suffices for graphs with negative edge weights [GT89] file:///E|/BOOK/BOOK4/NODE162.HTM (4 of 5) [19/1/2003 1:31:00]

Shortest Path

Related Problems: Network flow (see page

), motion planning (see page

).

Next: Transitive Closure and Reduction Up: Graph Problems: Polynomial-Time Previous: Minimum Spanning Tree Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE162.HTM (5 of 5) [19/1/2003 1:31:00]

Transitive Closure and Reduction

Next: Matching Up: Graph Problems: Polynomial-Time Previous: Shortest Path

Transitive Closure and Reduction

Input description: A directed graph G=(V,E). Problem description: For transitive closure, construct a graph G'=(V,E') with edge iff there is a directed path from i to j in G. For transitive reduction, construct a small graph G'=(V,E') with a directed path from i to j in G' iff . Discussion: Transitive closure can be thought of as establishing a data structure that makes it possible to solve reachability questions (can I get to x from y?) efficiently. After the preprocessing of constructing the transitive closure, all reachability queries can be answered in constant time by simply reporting a matrix entry. Transitive closure is fundamental in propagating the consequences of modified attributes of a graph G. For example, consider the graph underlying any spreadsheet model, where the vertices are cells and there is an edge from cell i to cell j if the result of cell j depends on cell i. When the value of a given cell is modified, the values of all reachable cells must also be updated. The identity of these cells is revealed by the transitive closure of G. Many database problems reduce to computing transitive closures, for analogous reasons. There are three basic algorithms for computing transitive closure:

file:///E|/BOOK/BOOK4/NODE163.HTM (1 of 4) [19/1/2003 1:31:01]

Transitive Closure and Reduction

●

The simplest algorithm just performs a breadth-first or depth-first search from each vertex and keeps track of all vertices encountered. Doing n such traversals gives an O(n (n+m) ) algorithm, which degenerates to cubic time if the graph is dense. This algorithm is easily implemented, runs well on sparse graphs, and is likely the right answer for your application. If the transitive closure of G will be dense, a better algorithm exploits the fact that the strongly

●

). All pairs of connected components of G can be computed in linear time (see Section vertices in each strongly connected component are mutually reachable. Further, if (x,y) is an edge between two vertices in different strongly connected components, every vertex in y's component is reachable from each vertex in x's component. Thus the problem reduces to finding the transitive closure on a graph of strongly connected components, which should have considerably fewer edges and vertices than G. Warshall's algorithm constructs transitive closures in with a simple, slick algorithm that is

●

identical to Floyd's all-pairs-shortest-path algorithm of Section . If we are interested only in the transitive closure, and not the length of the resulting paths, we can reduce storage by retaining only one bit for each matrix element. Thus iff j is reachable from i using only vertices as intermediates. Another related algorithm, discussed in the references, runs in the same time as matrix multiplication. You might conceivably win for large n by using Strassen's fast matrix multiplication algorithm, although I for one wouldn't bother trying. Since transitive closure is provably as hard as matrix multiplication, there is little hope for a significantly faster algorithm. Transitive reduction (also known as minimum equivalent digraph) is essentially the inverse operation of transitive closure, namely reducing the number of edges while maintaining identical reachability properties. The transitive closure of G is identical to the transitive closure of the transitive reduction of G. The primary application of transitive reduction is space minimization, by eliminating redundant edges from G that do not effect reachability. Transitive reduction also arises in graph drawing, where it is important to eliminate as many unnecessary edges as possible in order to reduce the visual clutter. Although the transitive closure of G is uniquely defined, a graph may have many different transitive reductions, including G itself. We want the smallest such reduction, but there can be multiple formulations of the problem: ●

●

A linear-time, quick-and-dirty transitive reduction algorithm identifies the strongly connected components of G, replaces each by a simple directed cycle, and adds these edges to those bridging the different components. Although this reduction is not provably minimal, it is likely to be pretty close on typical graphs. One catch with this heuristic is that it can add edges to the transitive reduction of G that are not in G. This may or may not be a problem for your application. If, in fact, all edges of the transitive reduction of G must be in G, we must abandon hope of

file:///E|/BOOK/BOOK4/NODE163.HTM (2 of 4) [19/1/2003 1:31:01]

Transitive Closure and Reduction

finding the minimum size reduction. To see why, consider a directed graph consisting of one strongly connected component, so that every vertex can reach every other vertex. The smallest possible transitive reduction will be a simple directed cycle, consisting of exactly n edges. This is possible as a subset of edges only if G is Hamiltonian, thus proving that finding the smallest subset of edges is NP-complete.

●

A heuristic for finding such a transitive reduction is to consider each edge successively and delete it if its removal does not change the transitive reduction. Implementing this efficiently means minimizing the time spent on reachability tests. Observe that directed edge (i,j) can be eliminated whenever there is another path from i to j avoiding this edge. If we are allowed to have arbitrary pairs of vertices as edges in the reduction and need the minimum size reduction, it can be found in time. See the references below for details. However, the quick-and-dirty algorithm above will likely suffice for most applications and will be easier to program as well as more efficient.

Implementations: LEDA (see Section using O(n m) time [GK79].

) provides an implementation of transitive closure in C++

Combinatorica [Ski90] provides Mathematica implementations of transitive closure and reduction, as well as the display of partial orders requiring transitive reduction. See Section

.

Notes: Van Leeuwen [vL90a] provides an excellent survey on transitive closure and reduction, including 33 references. Good expositions of Warshall's algorithm [War62] include [Baa88, CLR90, Man89]. The equivalence between matrix multiplication and transitive closure was proven by Fischer and Meyer [FM71], with good expositions including [AHU74]. reduction algorithm, was The equivalence between transitive closure and reduction, as well as the established in [AGU72]. Empirical studies of transitive closure algorithms include [SD75]. Estimating the size of the transitive closure is important in database query optimization. algorithm for estimating the size of the closure is given by Cohen [Coh94]. Related Problems: Connected components (see page

), shortest path (see page

Next: Matching Up: Graph Problems: Polynomial-Time Previous: Shortest Path file:///E|/BOOK/BOOK4/NODE163.HTM (3 of 4) [19/1/2003 1:31:01]

).

A linear-time

Transitive Closure and Reduction

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE163.HTM (4 of 4) [19/1/2003 1:31:01]

Matching

Next: Eulerian Cycle / Chinese Up: Graph Problems: Polynomial-Time Previous: Transitive Closure and Reduction

Matching

Input description: A (weighted) graph G=(V,E). Problem description: Find the largest-size set of edges S from E such that each vertex in V is incident to at most one edge of S. Discussion: Consider a set of employees, each of whom is capable of doing some subset of the tasks that must be performed. We seek to find an assignment of employees to tasks such that each task is assigned to a unique employee. Each mapping between an employee and a task they can handle defines an edge, so what we need is a set of edges with no employee or job in common, i.e. a matching. Efficient algorithms for constructing matchings are based on constructing augmenting paths in graphs. Given a (partial) matching M in a graph G, an augmenting path P is a path of edges where every oddnumbered edge (including the first and last edge) is not in M, while every even-numbered edge is. Further, the first and last vertices must not be already in M. By deleting the even-numbered edges of P from M and replacing them with the odd-numbered edges of P, we enlarge the size of the matching by one edge. Berge's theorem states that a matching is maximum if and only if it does not contain any augmenting path. Therefore, we can construct maximum-cardinality matchings by searching for file:///E|/BOOK/BOOK4/NODE164.HTM (1 of 4) [19/1/2003 1:31:03]

Matching

augmenting paths and stopping when none exist. This basic matching framework can be enhanced in several ways, while remaining essentially the same assignment problem: ●

●

●

Is your graph weighted or unweighted? - Many matching applications are based on unweighted graphs. Perhaps we seek to maximize the number of tasks performed, where each task is as good as another. Such a problem seeks a maximum cardinality matching. We say that a matching is perfect if every vertex is involved in the matching. For certain applications, we need to augment each edge with a weight, perhaps reflecting the salary of the employee or their effectiveness at a given task. The problem now becomes constructing a maximum weighted matching, i.e. the set of independent edges of maximum total cost. By setting all edge weights to be 1, any algorithm for finding weighted matchings can be used to solve maximum cardinality matching. What if certain employees can be given multiple jobs? - In a natural generalization of the assignment problem, certain employees can given more than one task to do. We do not seek a matching so much as a covering with small ``stars''. Such multiple jobs can be modeled by simply replicating the employee vertex as many times as the number of jobs she can handle. By adding sufficiently complicated constraints on the solution, we will eventually require the use of full integer programming. Is your graph bipartite? - Many common matching problems involve bipartite graphs, as in the classical assignment problem of jobs to workers. This is fortunate because faster and simpler algorithms exist for bipartite matching. General graphs prove trickier because it is possible to have augmenting paths that are odd-length cycles, i.e. the first and last vertices are the same. Such cycles (or blossoms) are impossible in bipartite graphs, which by definition do not contain odd-length cycles. The standard algorithms for bipartite matching are based on network flow, using a simple transformation to convert a bipartite graph into an equivalent flow graph.

Another common ``application'' of bipartite matching is in marrying off a set of boys to a set of girls such that each boy gets a girl he likes. This can be modeled as a bipartite matching problem, with an edge between any compatible boy and girl. This is possible only for graphs with perfect matchings. An interesting related problem seeks a matching such that no parties can be unhappy enough to seek to break the matching. That is, once each of the boys has ranked each of the girls in terms of desirability, and the girls do the same to the boys, we seek a matching with the property that there are no marriages of the form and , where and in fact prefer each other to their own spouses. In real life, these two would run off with each other, breaking the marriages. A marriage without any such couples is said to be stable. It is a surprising fact that no matter how the boys and girls rate each other, there is always at least one file:///E|/BOOK/BOOK4/NODE164.HTM (2 of 4) [19/1/2003 1:31:03]

Matching

stable marriage. Further, such a marriage can be found in time. An important application of stable marriage occurs in the annual matching of medical residents to hospitals. Implementations: The highest performance code available for constructing a maximum-cardinality bipartite matching of maximum weight in graphs is CSA [GK93], developed in the C language by Goldberg and Kennedy. This code is based on a cost-scaling network flow algorithm. They report solving instances with over 30,000 vertices in a few minutes on a Sun Sparc-2 workstation. Their codes are available for noncommercial use from http://www.neci.nj.nec.com/homepages/avg.html The First DIMACS Implementation Challenge [JM93] focused on network flows and matching. Several instance generators and implementations for maximum weight and maximum cardinality matching were collected, which can be obtained by anonymous ftp from dimacs.rutgers.edu in the directory pub/netflow/matching. These include: ●

●

●

A maximum-cardinality matching solver in Fortran 77 by R. Bruce Mattingly and Nathan P. Ritchey that seems capable of solving instances of 5,000 nodes and 60,000 edges in under 30 seconds. A maximum-cardinality matching solver in C by Edward Rothberg, that implements Gabow's algorithm. A maximum-weighted matching solver in C by Edward Rothberg. This is slower but more general than his unweighted solver described above. For example, it took over 30 seconds on a weighted graph with 500 nodes and 4,000 edges.

LEDA (see Section ) provides efficient implementations in C++ for both maximum cardinality and maximum weighted matching, on both bipartite and general graphs. Sedgewick [Sed92] provides a simple implementation of the stable marriage theorem in C++. See Section

for details.

Pascal implementations of maximum cardinality matching appears in [SDK83]. Alternative Pascal maximum-cardinality and bipartite matching codes appear in [MS91]. All are discussed in Section

.

) contains an implementation of the Hungarian algorithm for The Stanford GraphBase (see Section bipartite matching. To provide readily visualized weighted bipartite graphs, Knuth uses a digitized version of the Mona Lisa and seeks row/column disjoint pixels of maximum brightness. Matching is also used to construct clever, resampled ``domino portraits.'' Algorithm 548 [CT80] presents a Fortran code for the assignment problem. Algorithm 575 [Duf81] permutes a matrix so as to minimize the number of zeros along the diagonal, which involves solving a matching problem. Both codes are available from Netlib (see Section

file:///E|/BOOK/BOOK4/NODE164.HTM (3 of 4) [19/1/2003 1:31:03]

).

Matching

Combinatorica [Ski90] provides a (slow) Mathematica implementations of bipartite and maximal matching, as well as the stable marriage theorem. See Section

.

Notes: Lovász and Plummer [LP86] is the definitive reference on matching theory and algorithms. Survey articles on matching algorithms include [Gal86]. Good expositions on network flow algorithms for bipartite matching include [CLR90, Eve79a, Man89], and those on the Hungarian method include [Law76, PS82]. The best algorithm for maximum bipartite matching, due to Hopcroft and Karp [HK73], . repeatedly finds the shortest augmenting paths instead of using network flow, and runs in Expositions on the augmenting path method include [Man89, PS82, SDK83]. Edmond's algorithm [Edm65] for maximum-cardinality matching is of great historical interest for provoking questions about what problems can be solved in polynomial time. Expositions on Edmond's algorithm include [Law76, PS82, Tar83]. Gabow's [Gab76] implementation of Edmond's algorithm runs time. The best algorithm known for general matching runs in [MV80]. A faster in algorithm for matching in geometic graphs appears in [Vai88]. The theory of stable matching is thoroughly treated in [GI89]. The original algorithm for finding stable marriages is due to Gale and Shapely [GS62] with expositions including [Law76]. Related Problems: Eulerian cycle (see page

), network flow (see page

).

Next: Eulerian Cycle / Chinese Up: Graph Problems: Polynomial-Time Previous: Transitive Closure and Reduction Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE164.HTM (4 of 4) [19/1/2003 1:31:03]

Eulerian Cycle / Chinese Postman

Next: Edge and Vertex Connectivity Up: Graph Problems: Polynomial-Time Previous: Matching

Eulerian Cycle / Chinese Postman

Input description: A graph G=(V,E). Problem description: Find the shortest tour of G visiting each edge at least once. Discussion: Suppose you are given the map of a city and charged with designing the routes for garbage trucks, snow plows, or postmen. In all of these applications, every road in the city must be completely traversed at least once in order to ensure that all deliveries or pickups are made. For efficiency, you seek to minimize total drive time, or equivalently, the total distance or number of edges traversed. Such applications are variants of the Eulerian cycle problem, best characterized by the children's puzzle that asks them to draw a given figure completely without lifting their pencil off the paper and without repeating any edges. We seek a path or cycle through a graph that visits each edge exactly once. There are well-known conditions for determining whether a graph contains an Eulerian cycle, or path: ●

●

An undirected graph contains an Eulerian cycle iff (1) it is connected and (2) each vertex is of even degree. An undirected graph contains an Eulerian path iff (1) it is connected and (2) all but two vertices

file:///E|/BOOK/BOOK4/NODE165.HTM (1 of 4) [19/1/2003 1:31:06]

Eulerian Cycle / Chinese Postman

●

●

are of even degree. These two vertices will be the start and end points of the path. A directed graph contains an Eulerian cycle iff (1) it is connected and (2) each vertex has the same in-degree as out-degree. Finally, a directed graph contains an Eulerian path iff (1) it is connected and (2) all but two vertices have the same in-degree as out-degree, and these two vertices have their in-degree and out-degree differ by one.

Given this characterization of Eulerian graphs, it is easy to test in linear time whether such a cycle exists: test whether the graph is connected using DFS or BFS, and then count the number of odd-degree vertices. Actually constructing such a cycle also takes linear time. Use DFS to find a cycle in the graph. Delete this cycle and repeat until the entire set of edges has been partitioned into a set of edge-disjoint cycles. Since deleting a cycle reduces each vertex degree by an even number, the remaining graph will continue to satisfy the same Eulerian degree-bound conditions. For any connected graph, these cycles will have common vertices, and so by splicing these cycles in a ``figure eight'' at a shared vertex, we can construct a single circuit containing all of the edges. An Eulerian cycle, if one exists, solves the motivating snowplow problem, since any tour that visits each edge only once must have minimum length. However, it is unlikely that any real road network would happen to satisfy the degree conditions that make it Eulerian. We need to solve the more general Chinese postman problem, which minimizes the length of a cycle that traverses every edge at least once. In fact, it can be shown that this minimum cycle never visits any edge more than twice, so good tours exist for any road network. The optimal postman tour can be constructed by adding the appropriate edges to the graph G so as to make it Eulerian. Specifically, we find the shortest path between each pair of odd-degree vertices in G. Adding a path between two odd-degree vertices in G turns both of them to even-degree, thus moving us closer to an Eulerian graph. Finding the best set of shortest paths to add to G reduces to identifying a minimum-weight perfect matching in a graph on the odd-degree vertices, where the weight of edge (i,j) is the length of the shortest path from i to j. For directed graphs, this can be solved using bipartite matching, where the vertices are partitioned depending on whether they have more ingoing or outgoing edges. Once the graph is Eulerian, the actual cycle can be extracted in linear time using the procedure described above. Implementations: Unfortunately, we have not able to identify a suitable Chinese postman implementation. However, it should not be difficult for you to roll your own by using a matching code from Section algorithm.

and all-pairs shortest path from Section

. Matching is by far the hardest part of the

Combinatorica [Ski90] provides Mathematica implementations of Eulerian cycles and de Bruijn sequences.

See Section

.

file:///E|/BOOK/BOOK4/NODE165.HTM (2 of 4) [19/1/2003 1:31:06]

Eulerian Cycle / Chinese Postman

Nijenhuis and Wilf [NW78] provide an efficient Fortran routine to enumerate all Eulerian cycles of a graph by backtracking.

See Section

.

Notes: The history of graph theory began in 1736, when Euler [Eul36] first solved the seven bridges of Königsberg problem. Königsberg (now Kaliningrad) is a city on the banks of the Pregel river. In Euler's day there were seven bridges linking the banks and two islands, which can be modeled as a multigraph with seven edges and four vertices. Euler sought a way to walk over each of the bridges exactly once and return home, i.e. an Eulerian cycle. Since all four of the vertices had odd degree, Euler proved that such a tour is impossible. The bridges were destroyed in World War II. See [BLW76] for a translation of Euler's original paper and a history of the problem. Expositions on linear algorithms for constructing Eulerian cycles [Ebe88] include [Eve79a, Man89]. Fleury's algorithm [Luc91] is a direct and elegant approach to constructing Eulerian cycles. Start walking from any vertex, and erase any edge that has been traversed. The only criterion in picking the next edge is that we avoid using a bridge (edges whose deletion) unless there is no other alternative. No Eulerian graph contains a bridge, but what remains at some point on the walk ceases to be biconnected. The Euler's tour technique is an important paradigm in parallel graph algorithms. See [Man89] for an exposition. Efficient algorithms exist to count the number of Eulerian cycles in a graph [HP73]. The problem of finding the shortest tour traversing all edges in a graph was introduced by Kwan [Kwa62], hence the name Chinese postman. The bipartite matching algorithm for solving Chinese postman is due to Edmonds and Johnson [EJ73]. This algorithm works for both directed and undirected graphs, although the problem is NP-complete for mixed graphs [Pap76a]. Mixed graphs contain both directed and undirected edges. Expositions of the Chinese postman algorithm include [Law76]. A de Bruijn sequence S of span n on an alphabet of size is a circular string of length containing all , the strings of length n as substrings of S, each exactly once. For example, for n=3 and circular string 00011101 contains the following substrings in order: 000, 001, 011, 111, 110, 101, 010, 100. De Bruijn sequences can be thought of as ``safe cracker'' sequences, describing the shortest sequence of dial turns with positions sufficient to try out all combinations of length n. De Bruijn sequences can be constructed by building a graph whose vertices are all strings of length n-1, where there is an edge (u,v) iff and . Any Eulerian cycle on this graph describes a de Bruijn sequence. For expositions on de Bruijn sequences and their construction, see [Eve79a, Ski90]. Related Problems: Matching (see page

), Hamiltonian cycle (see page

file:///E|/BOOK/BOOK4/NODE165.HTM (3 of 4) [19/1/2003 1:31:06]

).

Eulerian Cycle / Chinese Postman

Next: Edge and Vertex Connectivity Up: Graph Problems: Polynomial-Time Previous: Matching Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE165.HTM (4 of 4) [19/1/2003 1:31:06]

Edge and Vertex Connectivity

Next: Network Flow Up: Graph Problems: Polynomial-Time Previous: Eulerian Cycle / Chinese

Edge and Vertex Connectivity

Input description: A graph G. Optionally, a pair of vertices and t. Problem description: What is the smallest subset of vertices (edges) whose deletion will disconnect G? Alternatively, what is the smallest subset of vertices (edges) that will separate from t? Discussion: Graph connectivity often arises in problems related to network reliability. In the context of telephone networks, the vertex connectivity is the smallest number of switching stations that a terrorist must bomb in order to separate the network, i.e. prevent two unbombed stations from talking to each other. The edge connectivity is the smallest number of wires that need to be cut to accomplish the same thing. One well-placed bomb or snipping the right pair of cables suffices to disconnect the network above. The edge (vertex) connectivity of a graph G is the smallest number of edge (vertex) deletions sufficient to disconnect G. There is a close relationship between the two quantities. The vertex connectivity is always no smaller than the edge connectivity, since deleting one vertex incident on each edge in a cut set succeeds in disconnecting the graph. Of course, smaller vertex subsets may be possible. The minimum vertex degree is an upper bound on both the edge and vertex connectivity, since deleting all its neighbors (or the edges to all its neighbors) disconnects the graph into one big and one single-vertex component.

file:///E|/BOOK/BOOK4/NODE166.HTM (1 of 4) [19/1/2003 1:31:08]

Edge and Vertex Connectivity

Several connectivity problems prove to be of interest: ●

●

●

Is the graph already disconnected? - The simplest connectivity problem is testing whether the graph is in fact connected. A simple depth-first or breadth-first search suffices to identify all . For directed graphs, the issue is whether components in linear time, as discussed in Section the graph is strongly connected, meaning there is a directed path between any pair of vertices. In a weakly connected graph, there may exist paths to nodes from which there is no way to return. What if I want to split the graph into equal-sized pieces? - Often, what is sought is not the smallest set of edges or vertices whose deletion will disconnect the graph, but a small set that breaks the graph into roughly equal-sized pieces. For example, suppose we want to break a computer program spread across several files into two maintainable units. Construct a graph where the vertices are subroutines, with an edge between any two subroutines that interact, say by one calling the other. We seek to partition the routines into equal-sized sets so that the fewest pairs of interacting routines are spread across the set. . Although the This is the graph partition problem, which is further discussed in Section problem is NP-complete, reasonable heuristics exist to solve it. Is there one weak link in my graph? - We say that G is biconnected if no single vertex deletion is sufficient to disconnect G. Any vertex that is such a weak point is called an articulation vertex. A bridge is the analogous concept for edges, meaning a single edge whose deletion disconnects the graph. The simplest algorithms for identifying articulation vertices (or bridges) would try deleting vertices (or edges) one by one, and then using DFS or BFS to test whether the resulting graph is still connected. More complicated but linear-time algorithms exist for both problems, based on

●

. depth-first search. Implementations are described below and in Section Are arbitrary cuts OK, or must I separate a given pair of vertices? - There are two flavors of the general connectivity problem. One asks for the smallest cutset for the graph, the other for the smallest set to separate from t. Any algorithm for (-t)-connectivity can be used with each of the n(n-1)/2 possible pairs of vertices to give an algorithm for general connectivity. Less obviously, n1 runs will suffice, since we know that vertex must end up in a different component from at least one of the other n-1 vertices in any cut set.

Both edge and vertex connectivity can be found using network flow techniques. Network flow, , interprets a weighted graph as a network of pipes where the maximum capacity discussed in Section of an edge is its weight, and seeks to maximize the flow between two given vertices of the graph. The in G is exactly the weight of the smallest set of edges to disconnect G with maximum flow between and in different components. Thus the edge connectivity can be found by minimizing the flow between and each of the n-1 other vertices in an unweighted graph G. Why? After deleting the minimum-edge cut set, must be separated from some other vertex.

file:///E|/BOOK/BOOK4/NODE166.HTM (2 of 4) [19/1/2003 1:31:08]

Edge and Vertex Connectivity

Vertex connectivity is characterized by Menger's theorem, which states that a graph is k-connected if and only if every pair of vertices is joined by at least k vertex-disjoint paths. Network flow can again be used to perform this calculation, since in an unweighted graph G a flow of k between a pair of vertices implies k edge-disjoint paths. We must construct a graph G' with the property that any set of edgedisjoint paths in G' corresponds to vertex-disjoint paths in G. This can be done by replacing each vertex of G with two vertices and , such that edge for all , and by replacing every by edges , in G'. Thus two edge-disjoint paths in G' correspond to edge vertex-disjoint paths in G, and as such, the minimum maximum-flow in G' gives the vertex connectivity of G. Implementations: Combinatorica [Ski90] provides Mathematica implementations of edge and vertex connectivity, as well as connected, biconnected, and strongly connected components with bridges and articulation vertices. See Section

.

LEDA does not currently seem to have biconnected components and bridges, but it does contain all the tools to implement connectivity algorithms, including network flow. Pascal implementations of biconnected and strongly connected components appear in [MS91]. See Section

for details.

The Stanford GraphBase (see Section connected components.

) contains routines to compute biconnected and strongly

Notes: Good expositions on the network-flow approach to edge and vertex connectivity include [Eve79a, Ski90]. The correctness of these algorithms is based on Menger's theorem [Men27] that the connectivity is determined by the number of edge and vertex disjoint paths separating a pair of vertices. The maximum-flow, minimum-cut theorem is due to Ford and Fulkerson [FF62]. Efficient randomized algorithms for computing graph connectivity have recently been developed by Karger. See Motwani and Raghavan [MR95] for an excellent treatment of randomized algorithms. is due to Matula [Mat87]. Faster kA nonflow-based algorithm for edge k-connectivity in connectivity algorithms are known for certain small values of k. All three-connected components of a suffices to test 4-connectivity [KR91]. graph can be generated in linear time [HT73a], while Related Problems: Connected components (see page (see page

).

file:///E|/BOOK/BOOK4/NODE166.HTM (3 of 4) [19/1/2003 1:31:08]

), network flow (see page

), graph partition

Edge and Vertex Connectivity

Next: Network Flow Up: Graph Problems: Polynomial-Time Previous: Eulerian Cycle / Chinese Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE166.HTM (4 of 4) [19/1/2003 1:31:08]

Network Flow

Next: Drawing Graphs Nicely Up: Graph Problems: Polynomial-Time Previous: Edge and Vertex Connectivity

Network Flow

Input description: A graph G, where each edge e=(i,j) has a capacity

. A source node and sink node t.

Problem description: What is the maximum flow you can route from to t while respecting the capacity constraint of each edge? Discussion: Applications of network flow go far beyond plumbing. Finding the most cost-effective way to ship goods between a set of factories and a set of stores defines a network flow problem, as do resource-allocation problems in communications networks and a variety of scheduling problems. The real power of network flow is that a surprising variety of linear programming problems that arise in practice can be modeled as network flow problems, and that special-purpose network flow algorithms can solve such problems much faster than general-purpose linear programming methods. Several of the graph problems we have discussed in this book can be modeled as network flow, including bipartite matching, shortest path, and edge/vertex connectivity. The key to exploiting this power is recognizing that your problem can be modeled as network flow. This is not easy, and it requires experience and study. My recommendation is to first construct a linear file:///E|/BOOK/BOOK4/NODE167.HTM (1 of 4) [19/1/2003 1:31:10]

Network Flow

programming model for your problem and then compare it with the linear program for minimum-cost flow on a directed network G=(V,E). Let be a variable accounting for the flow from vertex i through edge j. The flow through edge j is constrained by its capacity, so

Further, at each nonsource or sink vertex, as much flow comes in as goes out, so

where we seek the assignment that minimizes

where

is the cost of a unit of flow from i through j.

Special considerations include: ●

●

●

●

What if all my costs are identical? - Simpler and faster algorithms exist for solving the simple (as opposed to min-cost) maximum flow problem. This problem arises in many applications, including connectivity and bipartite matching. What if all arc capacities are identical, either 0 or 1? - Faster algorithms exist for 0-1 network flows. See the references for details. What if I have multiple sources and/or sinks? - Such problems can be handled by modifying the network to create a vertex to serve as a super-source that feeds all the sources and a super-sink that drains all the sinks. What if I have multiple types of material moving through the network? - In modeling a telecommunications network, every message has a given source and destination. Each destination needs to receive exactly those calls sent to it, not a given quantity of communication from arbitrary places. This can be modeled as a multicommodity flow problem, where each call defines a different commodity and we seek to satisfy all demands without exceeding the total capacity of any edge. Linear programming will suffice for multicommodity flow if fractional flows are permitted. Unfortunately, multicommodity integral flow is NP-complete, even with only two commodities.

Network flow algorithms can be complicated, and significant engineering is required to optimize performance. Thus we strongly recommend that you use an existing code if possible, rather than implement your own. Excellent codes are available and are described below. The two primary classes of algorithms are: file:///E|/BOOK/BOOK4/NODE167.HTM (2 of 4) [19/1/2003 1:31:10]

Network Flow

●

●

Augmenting path methods - These algorithms repeatedly find a path of positive capacity from source to sink and add it to the flow. It can be shown that the flow through a network of rational capacities is optimal if and only if it contains no augmenting path, and since each augmentation adds to the flow, we will eventually find the maximum. The difference between network flow algorithms is in how they select the augmenting path. If we are not careful, each augmenting path will add but a little to the total flow, and so the algorithm might take a long time to converge. Preflow-push methods - These algorithms push flows from one vertex to another, ignoring until the end the constraint that the in-flow must equal the out-flow at each vertex. Preflow-push methods prove faster than augmenting path methods, essentially because multiple paths can be augmented simultaneously. These algorithms are the method of choice and are implemented in the best codes described below.

Implementations: The highest-performance code available for solving maximum-flow in graphs is PRF [CG94], developed in the C language by Cherkassky and Goldberg. They report solving instances with over 250,000 vertices in under two minutes on a Sun Sparc-10 workstation. For minimum-cost maxflow, the highest-performance code available is CS [Gol92], capable of solving instances of over 30,000 vertices in a few minutes on Sun Sparc-2 workstations. Both of their codes are available by ftp for noncommercial use from http://www.neci.nj.nec.com/homepages/avg.html. The First DIMACS Implementation Challenge on Network Flows and Matching [JM93] collected several implementations and generators for network flow, which can be obtained by anonymous ftp from dimacs.rutgers.edu in the directory pub/netflow/maxflow. These include: ●

●

A preflow-push network flow implementation in C by Edward Rothberg. It took under a second on a test graph of 500 nodes and 4,000 edges, but over an hour with 5,000 nodes and 40,000 edges. An implementation of eleven network flow variants in C, including the older Dinic and Karzanov algorithms by Richard Anderson and Joao Setubal. On an instance of 8,000 vertices and 12,000 edges, all options finished within two seconds.

Nijenhuis and Wilf [NW78] provide a Fortran implementation of Karzanov's algorithm for network flow. See Section

. Fortran minimum-cost flow codes are given in [PGD82] and [KH80].

) provides C++ implementations of maximum-flow and minimum-cost max-flow LEDA (see Section algorithms. It also provides an implementation of minimum cut. Pascal implementations of max-flow and minimum-cost flow algorithms are provided in [SDK83]. Alternative Pascal max-flow implementations appear in [MS91]. For both codes, see Section

file:///E|/BOOK/BOOK4/NODE167.HTM (3 of 4) [19/1/2003 1:31:10]

.

Network Flow

Combinatorica [Ski90] provides a (slow) Mathematica implementation of network flow, with applications to connectivity testing and matching. See Section

.

Notes: The definitive book on network flows and its applications is [AMO93]. Good expositions on preflow-push algorithms [GT88] include [CLR90]. Older augmenting path algorithms are discussed in [Eve79a, Man89, PS82]. Expositions on min-cost flow include [Law76, PS82, SDK83]. Expositions on the hardness of multicommodity flow [Ita78] include [Eve79a]. Conventional wisdom holds that network flow should be computable in O(nm) time, and there has been steady progress in lowering the time complexity. See [AMO93] for a history of algorithms for the time [GT88]. problem. The fastest known general network flow algorithm runs in Empirical studies of minimum-cost flow algorithms include [GKK74, Gol92]. Although network flow can be used to find minimum cut sets in graphs, faster algorithms are available, including [SW94] and [MR95]. Related Problems: Linear programming (see page

), matching (see page

), connectivity (see page

).

Next: Drawing Graphs Nicely Up: Graph Problems: Polynomial-Time Previous: Edge and Vertex Connectivity Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE167.HTM (4 of 4) [19/1/2003 1:31:10]

Drawing Graphs Nicely

Next: Drawing Trees Up: Graph Problems: Polynomial-Time Previous: Network Flow

Drawing Graphs Nicely

Input description: A graph G. Problem description: Draw a graph G so as to accurately reflect its structure. Discussion: Drawing graphs nicely is a problem that constantly arises in applications, such as displaying file directory trees or circuit schematic diagrams. Yet it is inherently ill-defined. What exactly does nicely mean? We seek an algorithm that shows off the structure of the graph so the viewer can best understand it. We also seek a drawing that looks aesthetically pleasing. Unfortunately, these are ``soft'' criteria for which it is impossible to design an optimization algorithm. Indeed, it is possible to come up with two or more radically different drawings of certain graphs and have each be most appropriate in certain contexts. For example, page these is the ``right'' one?

contains three different drawings of the Petersen graph. Which of

Several ``hard'' criteria can partially measure the quality of a drawing: ●

●

Crossings - We seek a drawing with as few pairs of crossing edges as possible, since they are distracting. Area - We seek a drawing that uses as little paper as possible, while ensuring that no pair of

file:///E|/BOOK/BOOK4/NODE168.HTM (1 of 4) [19/1/2003 1:31:12]

Drawing Graphs Nicely

●

●

vertices are placed too close to each other. Edge length - We seek a drawing that avoids long edges, since they tend to obscure other features of the drawing. Aspect ratio - We seek a drawing whose aspect ratio (width/height) reflects that of the desired output medium (typically a computer screen at 4/3) as close as possible.

Unfortunately, these goals are mutually contradictory, and the problem of finding the best drawing under any nonempty subset of them will likely be NP-complete. Two final warnings before getting down to business. For graphs without inherent symmetries or structure to display, it is likely that no really nice drawing exists, especially for graphs with more than 10 to 15 vertices. Even when a large, dense graph has a natural drawing, the shear amount of ink needed to draw it can easily overwhelm any display. A drawing of the complete graph on 100 vertices, , contains pixel display works out to 200 pixels an edge. What approximately 5,000 edges, which on a can you hope to see except a black blob in the center of the screen? Once all this is understood, it must be admitted that certain graph drawing algorithms can be quite effective and fun to play with. To choose the right one, first ask yourself the following questions: ●

●

●

●

Must the edges be straight, or can I have curves and/or bends? - Straight-line drawing algorithms are simpler than those with polygonal lines, but to visualize complicated graphs such as circuit designs, orthogonal polyline drawings seem to work best. Orthogonal means that all lines must be drawn either horizontal or vertical, with no intermediate slopes. Polyline means that each graph edge is represented by a chain of straight-line segments, connected by vertices or bends. Can you build a natural, application-specific drawing algorithm? - If your graph represents a network of cities and roads, you are unlikely to find a better drawing than placing the vertices in the same position as the cities on a map. This same principle holds for many different applications. Is your graph either planar or a tree? - If so, use one of the special planar graph or tree drawing and . algorithms of Sections How fast must your algorithm be? - If it is being used for interactive update and display, your graph drawing algorithm had better be very fast. You are presumably limited to using incremental algorithms, which change the positions of the vertices only in the immediate neighborhood of the edited vertex. If you need to print a pretty picture for extended study, you can afford to be a little more extravagant.

As a first, quick-and-dirty drawing for most applications, I recommend simply spacing the vertices evenly on a circle, and then drawing the edges as straight lines between vertices. Such drawings are easy to program and fast to construct, and have the substantial advantage that no two edges can obscure each other, since no three vertices will be nearly collinear. As soon as you allow internal vertices into your drawing, such artifacts can be hard to avoid. An unexpected pleasure with circular drawings is the symmetry that is sometimes revealed because consecutive vertices appear in the order they were inserted into the graph. Simulated annealing can be used to permute the circular vertex order so as to minimize file:///E|/BOOK/BOOK4/NODE168.HTM (2 of 4) [19/1/2003 1:31:12]

Drawing Graphs Nicely

crossings or edge length, and thus significantly improve the drawing. A good, general-purpose heuristic for drawing graphs models the graph as a system of springs and then uses energy minimization to space the vertices. Let adjacent vertices attract each other with a force proportional to the logarithm of their separation, while all nonadjacent vertices repel each other with a force proportional to their separation distance. These weights provide incentive for all edges to be as short as possible, while spreading the vertices apart. The behavior of such a system can be approximated by determining the force acting on each vertex at a particular time and then moving each vertex a small amount in the appropriate direction. After several such iterations, the system should stabilize on a reasonable drawing. The input and output figures above demonstrate the effectiveness of the spring embedding on a particular small graph. If you need a polyline graph drawing algorithm, my recommendation is that you study several of the implementations presented below, particularly graphEd and GraphViz, and see whether one of them can do the job. You will have to do a significant amount of work before you can hope to develop a better algorithm. Once you have a graph drawn, this opens another can of worms, namely where to place the edge/vertex labels. We seek to position the labels very close to the edges or vertices they identify, and yet to place them such that they do not overlap each other or important graph features. Map labeling heuristics are described in [WW95]. Optimizing label placement can be shown to be an NP-complete problem, but heuristics related to bin packing (see Section

) can be effectively used.

Implementations: Georg Sander maintains a comprehensive WWW page on graph drawing at http://www.cs.uni-sb.de/RW/users/sander/html/gstools.html. This is well worth checking out and probably should be your first stop in hunting down programs for graph drawing. The best ftp-able package of graph drawing algorithms is GraphEd, by Michael Himsolt. GraphEd [Him94] is a powerful interactive editor that enables the user to construct and manipulate both directed and undirected graphs. It contains a variety of graph and tree drawing algorithms, including planar drawings, polyline drawings, upward drawings of directed acyclic graphs (DAGs), and spring embeddings, and allows variations in node, edge, and label styles. Sgraph is an interface to GraphEd to support user-specific extensions written in C. It includes a modest library of algorithms for planarity testing, maximum flow, matching, and connectivity testing. GraphEd can be obtained by anonymous ftp from forwiss.uni-passau.de (132.231.20.10) in directory /pub/local/graphed. GraphEd is free for noncommercial use. Graphlet is a more recent project by the same group, available at http://www.fmi.unipassau.de/Graphlet. GraphViz is a popular graph drawing program developed by Stephen North of Bell Laboratories. It represents edges as splines and can construct useful drawings of quite large and complicated graphs. I recommend it, even though licensing considerations make it impossible to include on the Algorithm Repository or CD-ROM. A noncommercial license is available from http://portal.research.bellfile:///E|/BOOK/BOOK4/NODE168.HTM (3 of 4) [19/1/2003 1:31:12]

Drawing Graphs Nicely

labs.com/orgs/ssr/book/reuse/. Combinatorica [Ski90] provides Mathematica implementations of several graph drawing algorithms, including circular, spring, and ranked embeddings. See Section Combinatorica.

for further information on

daVinci is a graph drawing and editing system whose layout algorithm seeks to minimize edge crossings and line bends, from Michael Froehlich at the University of Bremen. Information about daVinci is available from http://www.informatik.uni-bremen.de/ davinci. Binaries are available for a variety of UNIX workstations, although source code is not available. Notes: A significant community of researchers in graph drawing has emerged in recent years, fueled by or fueling an annual conference on graph drawing, the proceedings of which are published by SpringerVerlag's Lecture Notes in Computer Science series. Perusing a volume of the proceedings will provide a good view of the state of the art and of what kinds of ideas people are thinking about. The best reference available on graph drawing is the annotated bibliography on graph drawing algorithms by Giuseppe Di Battista, Peter Eades, and Roberto Tamassia [BETT94], which is also available from http://www.cs.brown.edu/ rt. See [BGL 95] for an experimental study of graph drawing algorithms. Although it is trivial to space n points evenly along the boundary of a circle, the problem is considerably more difficult on the surface of a sphere. Extensive tables of such spherical codes for in up to five dimensions have been construction by Sloane, Hardin, and Smith, and are available from netlib (see Section

) in att/math/sloane.

Related Problems: Drawing trees (see page

), planarity testing (see page

).

Next: Drawing Trees Up: Graph Problems: Polynomial-Time Previous: Network Flow Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE168.HTM (4 of 4) [19/1/2003 1:31:12]

Drawing Trees

Next: Planarity Detection and Embedding Up: Graph Problems: Polynomial-Time Previous: Drawing Graphs Nicely

Drawing Trees

Input description: A tree T, which is a graph without any cycles. Problem description: A nice drawing of the tree T. Discussion: There are as many reasons to want to draw trees as there are types of structures that trees represent. Consider software and debugging tools that illustrate the hierarchical structure of file system directories, or that trace the execution of a program through its subroutines. The primary issue in drawing trees is establishing whether you are drawing free or rooted trees: ●

●

Rooted trees define a hierarchical order, emanating from a single source node identified as the root. Any drawing must reflect this hierarchical structure, as well as any additional applicationdependent constraints on the order in which children must appear. For example, family trees are rooted, with sibling nodes typically drawn from left to right in the order of birth. Free trees do not encode any structure beyond their connection topology. For example, there is no root associated with the minimum spanning tree of any graph, so a hierarchical drawing can be misleading. Such free trees might well inherit their drawing from that of the full underlying graph,

file:///E|/BOOK/BOOK4/NODE169.HTM (1 of 3) [19/1/2003 1:31:13]

Drawing Trees

such as the map of the cities whose distances define the minimum spanning tree. Since trees are always planar graphs, they can and should be drawn such that no two edges cross. Any of the planar drawing algorithms of Section could be used to do so. However, such algorithms are overkill, because much simpler algorithms can be used to construct planar drawings of trees. The springembedding heuristics of Section many applications.

also work well on free trees, although they may be too slow for

The most natural tree-drawing algorithms work with rooted trees. However, they can be used equally well with free trees by selecting one vertex to serve as the root of the drawing. This faux-root can be selected arbitrarily, or, even better, by using a center vertex of the tree. A center vertex minimizes the maximum distance to other vertices. For trees, the center always consists of either one vertex or two adjacent vertices, so pick either one of them. Further, the center of a tree can be identified in linear time by repeatedly trimming all the leaves until only the center remains. Your two primary options for drawing rooted trees are ranked and radial embeddings: ●

●

Ranked embeddings - Place the root in the top center of your page, and then partition the page into the root-degree number of top-down strips. Deleting the root creates the root-degree number of subtrees, each of which is assigned to its own strip. Draw each subtree recursively, by placing its new root (the vertex adjacent to the old root) in the center of its strip a fixed distance from the top, with a line from old root to new root. The output figure above is a nicely ranked embedding of a balanced binary tree. Such ranked embeddings are particularly effective for rooted trees used to represent a hierarchy, be it a family tree, a data structure, or a corporate ladder. The top-down distance illustrates how far each node is from the root. Unfortunately, such a repeated subdivision eventually produces very narrow strips, until most of the vertices are crammed into a small region of the page. You should adjust the width of each strip to reflect the total number of nodes it will contain, or even better, the maximum number of nodes on a single level. Radial embeddings - A better way to draw free trees is with a radial embedding, where the root/center of the tree is placed in the center of the drawing. The space around this center vertex is divided into angular sectors for each subtree. Although the same problem of cramping will eventually occur, radial embeddings make better use of space than ranked embeddings and appear considerably more natural for free trees. The rankings of vertices in terms of distance from the center is illustrated by the concentric circles of vertices.

Implementations: Georg Sander maintains a comprehensive WWW page on graph drawing at http://www.cs.uni-sb.de/RW/users/sander/html/gstools.html. This should probably be your first stop in hunting down programs for tree drawing.

file:///E|/BOOK/BOOK4/NODE169.HTM (2 of 3) [19/1/2003 1:31:13]

Drawing Trees

The best FTP-able package of graph drawing algorithms is GraphEd, by Michael Himsolt ([email protected]). It contains a variety of graph and tree drawing algorithms and an interface to support user-specific extensions written in C. See Section and other graph drawing systems. Combinatorica [Ski90] provides Mathematica implementations including radial and rooted embeddings. See Section

for more details on GraphEd

of several tree drawing algorithms,

for further information on Combinatorica.

Notes: The best reference available on graph drawing is the annotated bibliography on graph drawing algorithms by Giuseppe Di Battista, Peter Eades, and Roberto Tamassia [BETT94], also available via http://www.cs.brown.edu/ rt. Heuristics for tree layout have been studied by several researchers [RT81, Vau80, WS79], although under certain aesthetic criteria the problem is NP-complete [SR83]. Related Problems: Drawing graphs (see page

), planar drawings (see page

).

Next: Planarity Detection and Embedding Up: Graph Problems: Polynomial-Time Previous: Drawing Graphs Nicely Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE169.HTM (3 of 3) [19/1/2003 1:31:13]

Planarity Detection and Embedding

Next: Graph Problems: Hard Problems Up: Graph Problems: Polynomial-Time Previous: Drawing Trees

Planarity Detection and Embedding

Input description: A graph G. Problem description: Can G be drawn in the plane such that no two edges cross? If so, produce such a drawing. Discussion: Planar drawings (or embeddings) make it easy to understand the structure of a given graph by eliminating crossing edges, which are often confused as additional vertices. Graphs arising in many applications, such as road networks or printed circuit boards, are naturally planar because they are defined by surface structures. Planar graphs have a variety of nice properties, which can be exploited to yield faster algorithms for many problems on planar graphs. Perhaps the most important property is that every planar graph is sparse. Euler's formula shows that for planar graph G=(V,E), , so every planar graph contains a linear number of edges, and further, every planar graph must contain a vertex of degree at most 5. Since every subgraph of a planar graph is planar, this means that there is always a sequence of low-degree vertices whose deletion from G eventually leaves the empty graph.

file:///E|/BOOK/BOOK4/NODE170.HTM (1 of 3) [19/1/2003 1:31:14]

Planarity Detection and Embedding

The study of planarity has motivated much of the development of graph theory. To get a better appreciation of the subtleties of planar drawings, the reader is urged to construct a planar (noncrossing) , the complete graph on five vertices with any single edge deleted. embedding for the graph Then construct such an embedding where all the edges are straight. Finally, attempt to do the same for itself. It is useful to distinguish the problem of planarity testing (does a graph have a planar drawing) from constructing planar embeddings (actually finding the drawing), although both can be done in linear time. Surprisingly, many efficient algorithms for planar graphs do not make use of the drawing but use the low-degree deletion sequence described above. Algorithms for planarity testing begin by embedding an arbitrary cycle from the graph in the plane and then considering additional paths in G between vertices on this cycle. Whenever two such paths cross, one must be drawn outside the cycle and one inside. When three such paths mutually cross, there is no way to resolve the problem, and so the graph cannot be planar. Linear-time algorithms for planarity detection are based on depth-first search, but they are subtle and complicated enough that you would be wise to use an existing implementation if you can. Such path-crossing algorithms can be used to construct a planar embedding by inserting the paths into the drawing one by one. Unfortunately, because they work in an incremental manner, nothing prevents them from inserting many vertices and edges into a relatively small area of the drawing. Such cramping makes the drawing ugly and hard to understand, and is a major problem with planar-embedding algorithms. More recently, algorithms have been devised that construct planar-grid embeddings, where each vertex lies on a grid. Thus no region can get too cramped and no edge can get too long. Still, the resulting drawings tend not to look as natural as one might hope. For nonplanar graphs, what is often sought is a drawing that minimizes the number of crossings. Unfortunately, computing the crossing number is NP-complete. A useful heuristic extracts a large planar subgraph of G, embeds this subgraph, and then inserts the remaining edges one by one so as to minimize the number of crossings. This won't do much for dense graphs, which are doomed to have a large number of crossings, but it will work well for graphs that are almost planar, such as road networks with overpasses or printed circuit boards with multiple layers. Large planar subgraphs can be found by modifying planarity-testing algorithms to delete troublemaking edges. Implementations: LEDA (see Section ) provides a nice set of data structures and algorithms to support working on planar subdivisions. Included are both linear-time planarity testing and constructing straight-line planar-grid embeddings. GraphEd includes an implementation of both planarity testing and planar graph layout. See Section for more details on GraphEd. Combinatorica [Ski90] provides a (slow) Mathematica implementation of

file:///E|/BOOK/BOOK4/NODE170.HTM (2 of 3) [19/1/2003 1:31:14]

Planarity Detection and Embedding

planarity testing. See Section

.

Notes: Kuratowski [Kur30] gave the first characterization of planar graphs, namely that they do not contain a subgraph homeomorphic to or . Thus if you are still working on the exercise to embed , now is an appropriate time to give it up. Fary's theorem [F48] states that every planar graph can be drawn such that each edge is straight. Hopcroft and Tarjan [HT74] gave the first linear-time algorithm for drawing graphs. Expositions on linear-time planarity testing include [Eve79a]. Nishizeki and Chiba [NC88] provide a good reference to the algorithmic theory of planar graphs. Efficient algorithms for planar grid embeddings were first developed by [dFPP88]. See [CHT90] for an algorithm to find the maximum planar subgraph of a nonplanar graph. Outerplanar graphs are those that can be drawn such that all vertices lie on the outer and face of the drawing. Such graphs can be characterized as having no subgraph homeomorphic to can be recognized and embedded in linear time. Related Problems: Graph partition (see page

), drawing trees (see page

).

Next: Graph Problems: Hard Problems Up: Graph Problems: Polynomial-Time Previous: Drawing Trees Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE170.HTM (3 of 3) [19/1/2003 1:31:14]

Graph Problems: Hard Problems

Next: Clique Up: A Catalog of Algorithmic Previous: Planarity Detection and Embedding

Graph Problems: Hard Problems A cynical view of graph algorithms is that ``everything we want to do is hard.'' Indeed, no polynomialtime algorithms are known for any of the problems in this section. Further, with the exception of graph isomorphism, all of them are provably NP-complete. The theory of NP-completeness demonstrates that if any NP-complete problem has a polynomial-time algorithm, then polynomial-time algorithms must exist for all NP-complete problems. This seems sufficiently preposterous that NP-completeness suffices as a de facto proof that no efficient worst-case algorithm exists for the given problem. Still, do not abandon hope if your problem resides in this chapter. For each of these problems, we provide a recommended line of attack, be it through combinatorial search, heuristics, or approximation algorithms. For every problem, there exist restricted input instances that are polynomial-time solvable, and if you are lucky, perhaps your data happens to fall into one of these classes. Hard problems require a different methodology to work with than polynomial-time problems, but with care they can usually be dealt with successfully. The following books will help you deal with NP-complete problems: ●

●

●

Garey and Johnson [GJ79] - This is the classic reference on the theory of NP-completeness. Most notably, it contains a concise catalog of over four hundred NP-complete problems, with associated references and comments. As soon as you begin to doubt the existence of an efficient algorithm for a given problem, browse through the catalog. Indeed, this is the book that I reach for most often. Hochbaum [Hoc96] - This book surveys the state of the art in approximation algorithms for NPcomplete problems. Approximation algorithms efficiently produce solutions to problems that are always provably close to optimal. Crescenzi and Kann - This compendium of approximation algorithms for optimization problems is available at http://www.nada.kth.se/nada/theory/problemlist.html and is the place to look first for a provably good heuristic for any given problem.

file:///E|/BOOK/BOOK4/NODE171.HTM (1 of 2) [19/1/2003 1:31:15]

Graph Problems: Hard Problems

● ● ● ● ● ● ● ● ● ● ●

Clique Independent Set Vertex Cover Traveling Salesman Problem Hamiltonian Cycle Graph Partition Vertex Coloring Edge Coloring Graph Isomorphism Steiner Tree Feedback Edge/Vertex Set

Next: Clique Up: A Catalog of Algorithmic Previous: Planarity Detection and Embedding Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE171.HTM (2 of 2) [19/1/2003 1:31:15]

Clique

Next: Independent Set Up: Graph Problems: Hard Problems Previous: Graph Problems: Hard Problems

Clique

Input description: A graph G=(V,E). Problem description: What is the largest

such that for all

,

?

Discussion: When I went to high school, everybody complained about the ``clique'', a group of friends who all hung around together and seemed to dominate everything social. Consider a graph whose vertices represent a set of people, with edges between any pair of people who are friends. Thus the clique in high school was in fact a clique in this friendship graph. Identifying ``clusters'' of related objects often reduces to finding large cliques in graphs. One example is in a program recently developed by the Internal Revenue Service (IRS) to detect organized tax fraud, where groups of phony tax returns are submitted in the hopes of getting undeserved refunds. The IRS constructs graphs with vertices corresponding to submitted tax forms and with edges between any two tax-forms that appear suspiciously similar. A large clique in this graph points to fraud. Since any edge in a graph represents a clique of two vertices, the challenge lies not in finding a clique, but in finding a large one. And it is indeed a challenge, for finding a maximum clique is NP-complete. To make matters worse, not only is no good approximation algorithm known, it is provably hard to file:///E|/BOOK/BOOK4/NODE172.HTM (1 of 3) [19/1/2003 1:31:17]

Clique

approximate even to within a factor of . Theoretically, clique is about as hard as a problem in this book can get. So what can we hope to do about it? ●

●

Will a maximal clique suffice? - A maximal clique is a subset of vertices, each pair of which defines an edge, that cannot be enlarged by adding any additional vertex. This doesn't mean that it has to be large relative to the largest possible clique, but it might be. To find a nice maximal clique, sort the vertices from highest degree to lowest degree, put the first vertex in the clique, and then test each of the other vertices in order to see whether it is adjacent to all the clique vertices thus far. If so, add it; if not, continue down the list. In O(n m) time you will have a maximal, and hopefully large, clique. An alternative approach would be to incorporate some randomness into your vertex ordering and accept the largest maximal clique you find after a certain number of trials. What if I am looking for a large, dense subgraph instead of a perfect clique? - Insisting on perfect cliques to define clusters in a graph can be risky, since the loss of a single edge due to error will eliminate that vertex from consideration. Instead, we should seek large dense subgraphs, i.e. subsets of vertices that contain a large number of edges between them. Cliques are, by definition, the densest subgraphs possible. A simple linear-time algorithm can be used to find the largest set of vertices whose induced (defined) subgraph has minimum vertex degree , beginning by repeatedly deleting all the vertices whose degree is less than k. This may reduce the degree of other vertices below k if they were adjacent to any deleted low-degree vertices. By repeating this process until all remaining vertices have degree , we eventually construct the largest dense subgraph. This algorithm can be implemented in O(n+m) time by using adjacency lists and the constant-width priority queue of

●

. By continuing to delete the lowest-degree vertices, we will eventually end up with a Section clique, which may or may not be large depending upon the graph. What if the graph is planar? - Planar graphs cannot have cliques of size larger than four, or else they cease to be planar. Since any edge defines a clique of size two, the only interesting cases are cliques of 3 and 4 vertices. Efficient algorithms to find such small cliques consider the vertices from lowest to highest degree. Any planar graph must contain a vertex of degree at most 5 (see ), so there is only a constant-sized neighborhood to check exhaustively for a clique. Section Once we finish with this vertex, we delete it to leave a smaller planar graph with at least one lowdegree vertex. Repeat until the graph is empty.

If you really need to find the largest clique in a graph, an exhaustive search via backtracking provides the only real solution. We search through all k-subsets of the vertices, pruning a subset as soon as it contains a vertex that is not connected to all the rest. We never need consider a subset of size larger than the highest vertex degree in the graph, since the maximum degree gives an upper bound on the size of the largest clique in the graph. Similarly, as soon as we discover a clique of size k, no vertex of degree can help find a larger clique. To speed our search, we should delete all such useless vertices from G.

file:///E|/BOOK/BOOK4/NODE172.HTM (2 of 3) [19/1/2003 1:31:17]

Clique

Heuristics for finding large cliques based on randomized techniques such as simulated annealing are likely to work reasonably well. Implementations: Programs for the closely related problems of finding cliques and independent sets were sought for the Second DIMACS Implementation Challenge, held in October 1993. Programs and data from the challenge are available by anonymous ftp from dimacs.rutgers.edu. Source codes are available under pub/challenge/graph and test data under pub/djs. In particular, two C language programs by David S. Johnson and David L. Applegate are available. The dfmax.c implements a simple-minded branch-and-bound algorithm similar to that of [CP90]. The dmclique.c uses a ``semi-exhaustive greedy'' scheme for finding large independent sets described in [JAMS91]. Performance data for both programs is available in files results.dfmax and results.dmclique in directories /pub/challenge/graph/benchmarks/clique and /pub/challenge/graph/benchmarks/volume. Combinatorica [Ski90] provides (slow) Mathematica implementations of cliques, independent sets, and vertex covers.

See Section

.

Notes: Good expositions of the proof that clique is NP-complete [Kar72] include [CLR90, GJ79, Man89]. It is also given in Section . This reduction established that clique, vertex cover, and independent set are very closely related problems, so heuristics and programs that solve one of them may also produce reasonable solutions for the other two. The linear-time algorithm for constructing maximal induced subgraphs is discussed in [Man89]. That is shown in [BGS95]. clique cannot be approximated to within a factor of Related Problems: Independent set (see page

), vertex cover (see page

).

Next: Independent Set Up: Graph Problems: Hard Problems Previous: Graph Problems: Hard Problems Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE172.HTM (3 of 3) [19/1/2003 1:31:17]

Independent Set

Next: Vertex Cover Up: Graph Problems: Hard Problems Previous: Clique

Independent Set

Input description: A graph G=(V,E). Problem description: What is the largest subset S of vertices of V such that no pair of vertices in S defines an edge of E between them? Discussion: The need to find large independent sets typically arises in dispersion problems, where we seek a set of mutually separated points. For example, suppose you are trying to identify locations for a new franchise service such that no two locations are close enough to compete with each other. Construct a graph where the vertices are possible locations, and add edges between any two locations deemed close enough to interfere. The maximum independent set gives you the maximum number of franchises you can sell without cannibalizing sales. Independent sets avoid conflicts between elements and hence arise often in coding theory and scheduling problems. Define a graph whose vertices represent the set of possible code words, and add edges between any two code words sufficiently similar to be confused due to noise. The maximum independent set of this graph defines the highest capacity code for the given communication channel. Independent set is closely related to two other NP-complete problems: file:///E|/BOOK/BOOK4/NODE173.HTM (1 of 3) [19/1/2003 1:31:18]

Independent Set

●

●

Vertex coloring - A coloring of a graph G is in fact a partitioning of the vertices of G into a small number of independent sets, since any two vertices of the same color cannot have an edge between them. In fact, most scheduling applications of independent set are really coloring problems, since all tasks eventually must be completed. Clique - The complement of a graph G = (V,E) is a graph G' = (V,E'), where iff (i,j) is not in E. In other words, we replace each edge by a nonedge and vica versa. The maximum independent set in G is exactly the maximum clique in G', so these problems are essentially identical. Algorithms and implementations in Section set.

can thus be easily used for independent

The simplest reasonable heuristic is to find the lowest-degree vertex, add it to the independent set, and delete it and all vertices adjacent to it. Repeating this process until the graph is empty gives a maximal independent set, in that it can't be made larger just by adding vertices. Using randomization or perhaps some exhaustive search to distinguish among the low-degree vertices might result in somewhat larger independent sets. The independent set problem is in some sense dual to the graph matching problem. The former asks for a large set of vertices with no edge in common, while the latter asks for a large set of edges with no vertex in common. This suggests trying to rephrase your problem as a matching problem, which can be computed in polynomial time, while the maximum independent set problem is NP-complete. The maximum independent set of a tree can be found in linear time by (1) stripping off the leaf nodes, (2) adding them to the independent set, (3) deleting the newly formed leaves, and then (4) repeating from the first step on the resulting tree until it is empty. Implementations: Programs for the closely related problems of finding cliques and independent sets were sought for the Second DIMACS Implementation Challenge, held in October 1993. Programs and data from the challenge are available by anonymous ftp from dimacs.rutgers.edu. Source codes are available under pub/challenge/graph and test data under pub/djs. See Section

.

Combinatorica [Ski90] provides (slow) Mathematica implementations of cliques, independent sets, and vertex covers.

See Section

for further information on Combinatorica.

Neural-network heuristics for vertex cover and related problems such as clique and vertex coloring have been implemented in C and Mathematica by Laura Sanchis and Arun Jagota, and are available in the algorithm repository http://www.cs.sunysb.edu/ algorith. Notes: Independent set remains NP-complete for planar cubic graphs [GJ79]. However, it can be solved efficiently for bipartite graphs [Law76].

file:///E|/BOOK/BOOK4/NODE173.HTM (2 of 3) [19/1/2003 1:31:18]

Independent Set

Related Problems: Clique (see page

), vertex coloring (see page

), vertex cover (see page

Next: Vertex Cover Up: Graph Problems: Hard Problems Previous: Clique Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE173.HTM (3 of 3) [19/1/2003 1:31:18]

).

Vertex Cover

Next: Traveling Salesman Problem Up: Graph Problems: Hard Problems Previous: Independent Set

Vertex Cover

Input description: A graph G=(V,E). Problem description: What is the smallest subset of vertex of S?

such that each

contains at least one

Discussion: Vertex cover is a special case of the more general set cover problem, which takes as input an arbitrary collection of subsets of the universal set . We seek the smallest subset of subsets from S whose union is U. Set cover arises in many applications, including Boolean logic minimization. See Section

for a discussion of set cover.

To turn vertex cover into a set cover problem, let U be the complete set of edges, and create to be the set of edges incident on vertex i. A set of vertices defines a vertex cover in graph G iff the corresponding subsets define a set cover in the given instance. However, since each edge can be in only two different subsets, vertex cover instances are simpler than general set cover. The primary reason for distinguishing between the two problems is that vertex cover is a relative lightweight among NP-complete problems, and so can be effectively solved.

file:///E|/BOOK/BOOK4/NODE174.HTM (1 of 3) [19/1/2003 1:31:19]

Vertex Cover

Vertex cover and independent set are very closely related graph problems. Since every edge in E is (by definition) incident on a vertex in a cover S, there can be no edge for which both endpoints are not in S. Thus V-S must be an independent set. Further, since minimizing S is the same as maximizing V-S, a minimum vertex cover defines a maximum independent set, and vice versa. This equivalence means that if you have a program that solves independent set, you can use it on your vertex cover problem. Having two ways of looking at it can be helpful if you expect that either the cover or independent set is likely to contain only a few vertices, for it might pay to search all possible pairs or triples of vertices if you think that it will pay off. The simplest heuristic for vertex cover selects the vertex with highest degree, adds it to the cover, deletes all adjacent edges, and then repeats until the graph is empty. With the right data structures, this can be done in linear time, and the cover you get ``usually'' should be ``pretty good''. However, for certain input graphs the resulting cover can be times worse than the optimal cover. Much better is the following approximation algorithm, discussed in Section , which always finds a vertex cover whose size is at most twice as large as optimal. Find a maximal matching in the graph, i.e. a set of edges no two of which share a vertex in common and that cannot be made larger by adding additional edges. Such a maximal matching can be built incrementally, by picking an arbitrary edge e in the graph, deleting any edge sharing a vertex with e, and repeating until the graph is out of edges. Taking both of the vertices for each edge in the matching gives us a vertex cover, since we only delete edges incident to one of these vertices and eventually delete all the edges. Because any vertex cover must contain at least one of the two vertices in each matching edge just to cover the matching, this cover must be at most twice as large as that of the minimum cover. This heuristic can be tweaked to make it perform somewhat better in practice, without losing the performance guarantee or costing too much extra time. We can select the matching edges so as to ``kill off'' as many edges as possible, which should reduce the size of the maximal matching and hence the number of pairs of vertices in the vertex cover. Also, some vertices selected for our cover may in fact not be necessary, since all of their incident edges could also have been covered using other selected vertices. By making a second pass through our cover, we can identify and delete these losers. If we are really lucky, we might halve the size of our cover using these techniques. A problem that might seem closely related to vertex cover is edge cover, which seeks the smallest set of edges such that each vertex is included in one of the edges. In fact, edge cover can be efficiently solved by finding a maximum cardinality matching in G (see Section account for the unmatched vertices.

) and then selecting arbitrary edges to

Implementations: Programs for the closely related problems of finding cliques and independent sets were sought for the Second DIMACS Implementation Challenge, held in October 1993. See Section for details.

file:///E|/BOOK/BOOK4/NODE174.HTM (2 of 3) [19/1/2003 1:31:19]

Vertex Cover

Combinatorica [Ski90] provides (slow) Mathematica implementations of cliques, independent sets, and vertex covers.

See Section

for further information on Combinatorica.

Neural-network heuristics for vertex cover and related problems such as clique and vertex coloring have been implemented in C and Mathematica by Laura Sanchis and Arun Jagota, and are available in the algorithm repository http://www.cs.sunysb.edu/ algorith. Notes: Good expositions of the proof that vertex-cover is NP-complete [Kar72] include [CLR90, GJ79, Man89]. Good expositions on the 2-approximation algorithm include [CLR90]. The example that the greedy algorithm can be as bad as times optimal is due to [Joh74] and is presented in [PS82]. Related Problems: Independent set (see page

), set cover (see page

).

Next: Traveling Salesman Problem Up: Graph Problems: Hard Problems Previous: Independent Set Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE174.HTM (3 of 3) [19/1/2003 1:31:19]

Traveling Salesman Problem

Next: Hamiltonian Cycle Up: Graph Problems: Hard Problems Previous: Vertex Cover

Traveling Salesman Problem

Input description: A weighted graph G. Problem description: Find the cycle of minimum cost that visits each of the vertices of G exactly once. Discussion: The traveling salesman problem is the most notorious NP-complete problem. This is a function of its general usefulness, and because it is easy to explain to the public at large. Imagine a traveling salesman who has to visit each of a given set of cities by car. What is the shortest route that will enable him to do so and return home, thus minimizing his total driving? Although the problem arises in transportation applications, its most important applications arise in optimizing the tool paths for manufacturing equipment. For example, consider a robot arm assigned to solder all the connections on a printed circuit board. The shortest tour that visits each solder point exactly once defines the most efficient path for the robot. A similar application arises in minimizing the amount of time taken by a graphics plotter to draw a given figure. Several issues arise in solving TSPs: ●

Is the graph unweighted? - If the graph is unweighted, or all the edges have one of two cost

file:///E|/BOOK/BOOK4/NODE175.HTM (1 of 5) [19/1/2003 1:31:21]

Traveling Salesman Problem

●

●

●

values, the problem reduces to finding a Hamiltonian cycle. See Section for a discussion of this problem. Are you given as input n points or a weighted graph? - Geometric points are often easier to work with than a graph representation, for several reasons. First, they define a complete graph, so there is never an issue of finding a tour, just a good one. Second, although we always could construct the complete distance graph on the points and feed it to a graph solver, it might be more ) and work primarily from efficient to construct a sparse nearest-neighbor graph (see Section that. Finally, geometric instances inherently satisfy the triangle inequality, discussed below. How important is the restriction against visiting each vertex more than once? - The restriction that the tour not revisit any vertex may be important in certain applications, but it often is irrelevant. For example, the cheapest way to visit all vertices might involve repeatedly visiting a hub site, as is common in modern air travel. This issue does not arise whenever the graph observes the triangle inequality; that is, for all vertices , . In graphs that observe the triangle inequality, the shortest tour visits each vertex once. Heuristics work much better on graphs that do obey the triangle inequality. How important is that that you find the optimal tour? - If you insist on solving your TSP to optimality (and you probably shouldn't bother), there are two common approaches. Cutting plane methods model the problem as an integer program, then solve the linear programming relaxation of it. If the optimal solution is not at integer points, additional constraints designed to force integrality are added. Branch-and-bound algorithms perform a combinatorial search while maintaining careful upper and lower bounds on the cost of a tour or partial tour. In the hands of professionals, problems with thousands of vertices can be solved. In the hands of one gleaning their knowledge from this book, problems with 50 to maybe 100 vertices are potentially solvable, using the implementations discussed below.

Almost any flavor of TSP is going to be NP-complete, so the right way to proceed is with heuristics. These are often quite successful, typically coming within a few percent of the optimal solution, which is close enough for engineering work. Unfortunately, there have been literally dozens of heuristics proposed for TSPs, so the situation can be confusing. Empirical results in the literature are sometime contradictory. However, we recommend choosing from among the following heuristics: ●

Minimum spanning trees - A simple and popular heuristic, especially when the sites represent points in the plane, is based on the minimum spanning tree of the points. By doing a depth-first search of this tree, we walk over each edge of the tree exactly twice, once going down when we discover the new vertex and once going up when we backtrack. We can then define a tour of the vertices according to the order in which they were discovered and use the shortest path between each neighboring pair of vertices in this order to connect them. This path must be a single edge if the graph is complete and obeys the triangle inequality, as with points in the plane. As discussed in Section

, the resulting tour is always at most twice the length of the minimum TSP tour. In

file:///E|/BOOK/BOOK4/NODE175.HTM (2 of 5) [19/1/2003 1:31:21]

Traveling Salesman Problem

practice, it is usually better, typically 15% to 20% over optimal. Further, the time of the algorithm is bounded by that of computing the minimum spanning tree, only in the case of points in ●

●

the plane (see Section ). Incremental insertion methods - A different class of heuristics inserts new points into a partial tour one at a time (starting from a single vertex) until the tour is complete. The version of this heuristic that seems to work best is furthest point insertion: of all remaining points, insert the point v into partial tour T such that

The minimum ensures that we insert the vertex in the position that adds the smallest amount of distance to the tour, while the maximum ensures that we pick the worst such vertex first. This seems to work well because it first ``roughs out'' a partial tour before filling in details. Typically, such tours are only 5% to 10% longer than optimal. k-optimal tours - Substantially more powerful are the Kernighan-Lin, or k-opt class of heuristics. Starting from an arbitrary tour, the method applies local refinements to the tour in the hopes of improving it. In particular, subsets of edges are deleted from the tour and the k remaining subchains rewired in a different way to see if the resulting tour is an improvement. A tour is koptimal when no subset of k edges can be deleted and rewired so as to reduce the cost of the tour. Extensive experiments suggest that 3-optimal tours are usually within a few percent of the cost of optimal tours. For k > 3, the computation time increases considerably faster than solution quality. Two-opting a tour is a fast and effective way to improve any other heuristic. Simulated annealing provides an alternate mechanism to employ edge flips to improve heuristic tours.

Implementations: The world-record-setting traveling salesman program is by Applegate, Bixby, Chvatal, and Cook [ABCC95], which has solved instances as large as 7,397 vertices to optimality. At this time, the program is not being distributed. However, the authors seem willing to use it to solve TSPs sent to them. In their paper, they describe this work as neither theory nor practice, but sport - an almost recreational endeavor designed principally to break records. It is a very impressive piece of work, however. The TSPLIB library of test instances for the traveling salesman problem is available from Netlib, and by anonymous ftp from softlib.cs.rice.edu. See Section

.

Tsp_solve is a C++ code by Chad Hurwitz and Robert Craig that provides both heuristic and optimal solutions. Geometric problems of size up to 100 points are manageable. It is available from http://www.cs.sunysb.edu/ algorith or by e-mailing Chad Hurrwitz at [email protected]. A heuristic Euclidean TSP solver in C due to Lionnel Maugis is available from http://www.cenaath.cena.dgac.fr/ maugis/tsp.shar.

file:///E|/BOOK/BOOK4/NODE175.HTM (3 of 5) [19/1/2003 1:31:21]

Traveling Salesman Problem

Pascal implementations of branch-and-bound search and the insertion and Kerighan-Lin heuristics (for 2opt and 3-opt) appear in [SDK83]. For details, see Section

.

Algorithm 608 [Wes83] of the Collected Algorithms of the ACM is a Fortran implementation of a heuristic for the quadratic assignment problem, a more general problem that includes the traveling salesman as a special case. Algorithm 750 [CDT95] is a Fortran code for the exact solution of asymmetric TSP instances. See Section

for details.

) includes animations of both the minimum spanning tree heuristic and a genetic XTango (see Section algorithm for TSP. The latter converges sufficiently slowly to kill one's interest in genetic algorithms. Combinatorica [Ski90] provides (slow) Mathematica implementations of exact and approximate TSP solutions. See Section

.

Notes: The definitive reference on the traveling salesman problem is the book by Lawler et. al. [LLKS85]. Experimental results on heuristic methods for solving large TSPs include [Ben92a, GBDS80, Rei94]. Typically, it is possible to get within a few percent of optimal with such methods. TSPLIB [Rei91] provides the standard collection of hard instances of TSPs that arise in practice. The Christofides heuristic is an improvement of the minimum-spanning tree heuristic and guarantees a , where the bottleneck tour whose cost is at most 3/2 times optimal on Euclidean graphs. It runs in is the time it takes to find a minimum-weight perfect matching (see Section ). Good expositions of the Christofides heuristic [Chr76] include [Man89, PS85]. Expositions of the minimum spanning tree heuristic [RSL77] include [CLR90, O'R94, PS85]. Polynomial-time approximation schemes for Euclidean TSP have been recently developed by Arora [Aro96] and Mitchell [Mit96], which offer factor approximations in polynomial time for any They are of great theoretical interest, although any practical consequences remain to be determined. The history of progress on optimal TSP solutions is somewhat inspiring. In 1954, Dantzig, Fulkerson, and Johnson solved a symmetric TSP instance of 42 United States cities [DFJ54]. In 1980, Padberg and Hong solved an instance on 318 vertices [PH80]. Applegate et. al. [ABCC95] have recently solved problems that are twenty times larger than this. Some of this increase is due to improved hardware, but most is due to better algorithms. The rate of growth demonstrates that exact solutions to NP-complete problems can be obtained for large instances if the stakes are high enough. Unfortunately, they seldom are. Good expositions on branch-and-bound methods include [PS82, SDK83]. Good expositions of the Kernighan-Lin heuristic [LK73, Lin65] include [MS91, PS82, SDK83].

file:///E|/BOOK/BOOK4/NODE175.HTM (4 of 5) [19/1/2003 1:31:21]

.

Traveling Salesman Problem

Size is not the only criterion for hardness. One can easily construct an enormous graph consisting of one cheap cycle, for which it would be easy to find the optimal solution. For sets of points in convex position in the plane, the minimum TSP tour is described by its convex hull (see Section computed in time. Other easy special cases are known. Related Problems: Hamiltonian cycle (see page hull (see page

), which can be

), minimum spanning tree (see page

).

Next: Hamiltonian Cycle Up: Graph Problems: Hard Problems Previous: Vertex Cover Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE175.HTM (5 of 5) [19/1/2003 1:31:21]

), convex

Hamiltonian Cycle

Next: Graph Partition Up: Graph Problems: Hard Problems Previous: Traveling Salesman Problem

Hamiltonian Cycle

Input description: A graph G = (V,E). Problem description: Find an ordering of the vertices such that each vertex is visited exactly once. Discussion: The problem of finding a Hamiltonian cycle or path in a graph is a special case of the traveling salesman problem, one where each pair of vertices with an edge between them is considered to have distance 1, while nonedge vertex pairs are separated by distance . Closely related is the problem of finding the longest path or cycle in a graph, which occasionally arises in pattern recognition problems. Let the vertices in the graph correspond to possible symbols, and let edges link symbols that can possibly be next to each other. The longest path through this graph is likely the correct interpretation. The problems of finding longest cycles and paths are both NP-complete, even on very restrictive classes of unweighted graphs. There are several possible lines of attack, however: ●

Do you have to visit all the vertices or all the edges? - First verify that you really have a Hamiltonian cycle problem. As discussed in Section

file:///E|/BOOK/BOOK4/NODE176.HTM (1 of 3) [19/1/2003 1:31:23]

, fast algorithms exist for edge-tour, or

Hamiltonian Cycle

Eulerian cycle, problems, where you must visit all the edges without repetition. With a little cleverness, it is sometimes possible to reformulate a Hamiltonian cycle problem in terms of Eulerian cycles. Perhaps the most famous such instance is the problem of constructing de Bruijn ●

●

●

. sequences, discussed in Section Is there a serious penalty for visiting vertices more than once? - By phrasing the problem as minimizing the total number of vertices visited on a complete tour, we have an optimization problem that now allows room for heuristics and approximation algorithms. For example, , finding a spanning tree of the graph and doing a depth-first search, as discussed in Section yields a tour with at most 2n vertices. Using randomization or simulated annealing might bring the size of this down considerably. Am I seeking the longest path in a directed acyclic graph (DAG)? - The problem of finding the longest path in a DAG can be solved in linear time using dynamic programming. This is about the only interesting case of longest path for which efficient algorithms exist. Is my graph dense? - For sufficiently dense graphs, there always exists at least one Hamiltonian cycle, and further, such a cycle can be found quickly. An efficient algorithm for finding a Hamiltonian cycle in a graph where all vertices have degree is given in [Man89].

If you really must know whether your graph is Hamiltonian, backtracking with pruning is your only possible solution. Before you search, it pays to check whether your graph is biconnected (see Section ). If not, this means that the graph has an articulation vertex whose deletion will disconnect the graph and so cannot be Hamiltonian. Implementations: The football program of the Stanford GraphBase (see Section ) uses a stratified greedy algorithm to solve the asymmetric longest path problem. The goal is to derive a chain of football scores in order to establish the superiority of one football team over another. After all, if Virginia beat Illinois by 30 points, and Illinois beat Stony Brook by 14 points, then by transitivity Virginia would beat Stony Brook by 44 points if they played, right? We seek the longest path in a graph where the weight of an edge (x,y) is the number of points x beat y by. Nijenhuis and Wilf [NW78] provide an efficient routine to enumerate all Hamiltonian cycles of a graph by backtracking. See Section . Algorithm 595 [Mar83] of the Collected Algorithms of the ACM is a similar Fortran code that can be used as either an exact procedure or a heuristic by controlling the amount of backtracking. See Section

.

) is an algorithm animation system for UNIX and X-windows, which includes XTango (see Section an animation of a backtracking solution to the knight's tour problem. Combinatorica [Ski90] provides a Mathematica backtracking implementation of Hamiltonian cycle.

file:///E|/BOOK/BOOK4/NODE176.HTM (2 of 3) [19/1/2003 1:31:23]

Hamiltonian Cycle

See Section

.

Notes: Hamiltonian cycles - circuits that visit each vertex of a graph exactly once - apparently first arose in Euler's study of the knight's tour problem, although they were popularized by Hamilton's ``Around the World'' game in 1839. Good expositions of the proof that Hamiltonian cycle is NP-complete [Kar72] include [Baa88, CLR90, GJ79]. Techniques for solving optimization problems in the laboratory using biological processes have recently attracted considerable attention. In the original application of these ``biocomputing'' techniques, Adleman [Adl94b] solved a seven-vertex instance of the directed Hamiltonian path problem. Unfortunately, this approach requires an exponential number of molecules, and Avogadro's number . implies that such experiments are inconceivable for graphs beyond Related Problems: Eulerian cycle (see page

), traveling salesman (see page

).

Next: Graph Partition Up: Graph Problems: Hard Problems Previous: Traveling Salesman Problem Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE176.HTM (3 of 3) [19/1/2003 1:31:23]

Graph Partition

Next: Vertex Coloring Up: Graph Problems: Hard Problems Previous: Hamiltonian Cycle

Graph Partition

Input description: A (weighted) graph G=(V,E) and integers j, k, and m. Problem description: Partition the vertices into m subsets such that each subset has size at most j, while the cost of the edges spanning the subsets is bounded by k. Discussion: Graph partitioning arises as a preprocessing step to divide-and-conquer algorithms, where it is often a good idea to break things into roughly equal-sized pieces. It also arises when dealing with extremely large graphs, when we need to cluster the vertices into logical components for storage (to improve virtual memory performance) or for drawing purposes (to collapse dense subgraphs into single nodes in order to reduce cluttering). Several different flavors of graph partitioning arise depending on the desired objective function: ●

Minimum cut set - The smallest set of edges to cut that will disconnect a graph can be efficiently

●

for more on connectivity algorithms. Since found using network flow methods. See Section the smallest cutset can split off only a single vertex, the resulting partition might be very unbalanced. Hence ... Graph partition - A better partition criterion seeks a small cut that partitions the vertices into

file:///E|/BOOK/BOOK4/NODE177.HTM (1 of 3) [19/1/2003 1:31:24]

Graph Partition

roughly equal-sized pieces. Unfortunately, this problem is NP-complete. Fortunately, heuristics discussed below work well in practice. Certain special graphs always have small separators, which partition the vertices into balanced pieces. For any tree, there always exists a single vertex whose deletion partitions the tree so that no component contains more than n/2 of the original n vertices. These components need not always be connected. For example, consider the separating vertex of a star-shaped tree. However, the separating vertex can be found in linear time using depth first-search. Similarly, every planar graph has a set of vertices whose deletion leaves no component with more than 2n/3 vertices. Such separators provide a particularly useful way to decompose planar graphs.

Figure: The maximum cut of a graph ●

Maximum cut - Given an electronic circuit specified by a graph, the maximum cut defines the largest amount of data communication that can simultaneously take place in the circuit. The highest-speed communications channel should thus span the vertex partition defined by the maximum edge cut. Finding the maximum cut in a graph is NP-complete, despite the existence of algorithms for min-cut. However, heuristics similar to those of graph partitioning can work well.

The basic approach to dealing with graph partitioning or max-cut problems is to construct an initial partition of the vertices (either randomly or according to some problem-specific strategy) and then sweep through the vertices, deciding whether the size of the cut would increase or decrease if we moved this vertex over to the other side. The decision to move v can be made in time proportional to its degree by simply counting whether more of v's neighbors are on the same team as v or not. Of course, the desirable side for v will change if many of its neighbors jump, so multiple passes are likely to be needed before the process converges on a local optimum. Even such a local optimum can be arbitrarily far away from the global max-cut. There are many variations of this basic procedure, by changing the order we test the vertices in or moving clusters of vertices simultaneously. Using some form of randomization, particularly simulated file:///E|/BOOK/BOOK4/NODE177.HTM (2 of 3) [19/1/2003 1:31:24]

Graph Partition

annealing, is almost certain to be a good idea. Implementations: Jon Berry's implementations of several graph partitioning heuristics, including Kernighan-Lin, simulated annealing, and path optimization are available from http://www.elon.edu/users/f/berryj/www/. A non-network-flow-based implementation of minimum cut is included with LEDA (see Section

).

Notes: The fundamental heuristic for graph partitioning is due to Kernighan and Lin [KL70]. Empirical results on graph partitioning heuristics include [BG95, LR93]. The planar separator theorem and an efficient algorithm for finding such a separator are due to Lipton and Tarjan [LT79, LT80]. Although network flow can be used to find minimum cut sets in graphs, faster algorithms are available, including [SW94] and [Kar96a]. Expositions on the hardness of max-cut [Kar72] include [Eve79a]. Note that any random vertex partition will expect to cut half of the edges in the graph, since the probability that the two vertices defining an edge end up on different sides of the partition is 1/2. Goemans and Williamson [GW95] gave an 0.878factor approximation algorithm for maximum-cut, based on semidefinite programming techniques. Tighter analysis of this algorithm followed by Karloff [Kar96b]. Related Problems: Edge/vertex connectivity (see page

), network flow (see page

).

Next: Vertex Coloring Up: Graph Problems: Hard Problems Previous: Hamiltonian Cycle Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE177.HTM (3 of 3) [19/1/2003 1:31:24]

Vertex Coloring

Next: Edge Coloring Up: Graph Problems: Hard Problems Previous: Graph Partition

Vertex Coloring

Input description: A graph G=(V,E). Problem description: Color the vertices of V using the minimum number of colors such that for each edge , vertices i and j have different colors. Discussion: Vertex coloring arises in a variety of scheduling and clustering applications. Compiler optimization is the canonical application for coloring, where we seek to schedule the use of a finite number of registers. In a program fragment to be optimized, each variable has a range of times during which its value must be kept intact, in particular, after it is initialized and before its final use. Any two variables whose life spans intersect cannot be placed in the same register. Construct a graph where there is a variable associated with each vertex and add an edge between any two vertices whose variable life spans intersect. A coloring of the vertices of this graph assigns the variables to classes such that two variables with the same color do not clash and so can be assigned to the same register. No conflicts can occur if each vertex is colored with a distinct color. However, our goal is to find a coloring using the minimum number of colors, because computers have a limited number of registers. The smallest number of colors sufficient to vertex color a graph is known as its chromatic number.

file:///E|/BOOK/BOOK4/NODE178.HTM (1 of 4) [19/1/2003 1:31:26]

Vertex Coloring

Several special cases of interest arise in practice: ●

Can I color the graph using only two colors? - An important special case is testing whether a graph is bipartite, meaning it can be colored using two different colors. Such a coloring of the vertices of a bipartite graph means that the graph can be drawn with the red vertices on the left and the blue vertices on the right such that all edges go from left to right. Bipartite graphs are fairly simple, yet they arise naturally in such applications as mapping workers to possible jobs. Fast, simple algorithms exist for problems such as matching (see Section

●

●

) on bipartite graphs.

Testing whether a graph is bipartite is easy. Color the first vertex blue, and then do a depth-first search of the graph. Whenever we discover a new, uncolored vertex, color it opposite that of its parent, since the same color would cause a clash. If we ever find an edge where both vertices have been colored identically, then the graph cannot be bipartite. Otherwise, this coloring will be a 2coloring, and it is constructed in O(n+m) time. Is the graph planar, or are all vertices of low degree? - The famous 4-color theorem states that every planar graph can be vertex colored using at most 4 distinct colors. Efficient algorithms for finding a 4-coloring are known, although it is NP-complete to decide whether a given planar graph is 3-colorable. There is a very simple algorithm that finds a vertex coloring of any planar graph using at most 6 colors. In any planar graph, there exists a vertex of degree at most five. Delete this vertex and recursively color the graph. This vertex has at most five neighbors, which means that it can always be colored using one of the six colors that does not appear as a neighbor. This works because deleting a vertex from a planar graph leaves a planar graph, so we always must have a low-degree vertex to delete. The same idea can be used to color any graph of maximum degree using colors in time. Is this an edge coloring problem? - Certain vertex coloring problems can be modeled as edge coloring, where we seek to color the edges of a graph G such that no two edges with a vertex in common are colored the same. The payoff is that there is an efficient algorithm that always returns a near-optimal edge coloring. Algorithms for edge coloring are the focus of Section

.

Computing the chromatic number of a graph is NP-complete, so if you need an exact solution you must resort to backtracking, which can be surprisingly effective in coloring certain random graphs. It remains hard to compute a provably good approximation to the optimal coloring, so expect no guarantees. Incremental methods appear to be the heuristic of choice for vertex coloring. As in the previous algorithm for planar graphs, vertices are colored sequentially, with the colors chosen in response to colors already assigned in the vertex's neighborhood. These methods vary in how the next vertex is selected and how it is assigned a color. Experience suggests inserting the vertices in nonincreasing order of degree, since high-degree vertices have more color constraints and so are most likely to require an

file:///E|/BOOK/BOOK4/NODE178.HTM (2 of 4) [19/1/2003 1:31:26]

Vertex Coloring

additional color if inserted late. Incremental methods can be further improved by using color interchange. Observe that taking a properly colored graph and exchanging two of the colors (painting the red vertices blue and the blue vertices red) leaves a proper vertex coloring. Now suppose we take a properly colored graph and delete all but the red and blue vertices. If the remaining graph (the induced subgraph) consists of two or more connected components, we can repaint one or more of the components, again leaving a proper coloring. After such a recoloring, some vertex v previously adjacent to both red and blue vertices might now be only adjacent to blue vertices, thus freeing v to be colored red. Color interchange is a win in terms of producing better colorings, at a cost of increased time and implementation complexity. Implementations are described below. Simulated annealing algorithms that incorporate color interchange to move from state to state are likely to be even more effective. Implementations: Graph coloring has been blessed with two distinct and useful WWW resources. Michael Trick's page, http://mat.gsia.cmu.edu/COLOR/color.html, provides a nice overview of applications of graph coloring, an annotated bibliography, and a collection of over seventy graph coloring instances arising in applications such as register allocation and printed circuit board testing. Finally, it contains a C language implementation of an exact coloring algorithm, DSATUR. Joseph C. Culberson's WWW page on graph coloring, http://web.cs.ualberta.ca/ joe/Coloring/, provides an extensive bibliography and a collection of programs to generate hard graph coloring instances. Programs for the closely related problems of finding cliques and vertex coloring graphs were sought for the Second DIMACS Implementation Challenge, held in October 1993. Programs and data from the challenge are available by anonymous ftp from dimacs.rutgers.edu. Source codes are available under pub/challenge/graph and test data under pub/djs, including a simple ``semi-exhaustive greedy'' scheme used in the graph coloring algorithm XRLF [JAMS91]. Pascal implementations of backtracking algorithms for vertex coloring and several heuristics, including largest-first and smallest-last incremental orderings and color interchange, appear in [SDK83]. See Section

.

XTango (see Section ) is an algorithm animation system for UNIX and X-windows, and includes an animation of vertex coloring via backtracking. Nijenhuis and Wilf [NW78] provide an efficient Fortran implementation of chromatic polynomials and vertex coloring by backtracking. See Section

.

Combinatorica [Ski90] provides Mathematica implementations of bipartite graph testing, heuristic colorings, chromatic polynomials, and vertex coloring by backtracking. See Section file:///E|/BOOK/BOOK4/NODE178.HTM (3 of 4) [19/1/2003 1:31:26]

.

Vertex Coloring

Notes: An excellent source on vertex coloring heuristics is Syslo, Deo, and Kowalik [SDK83], which includes experimental results. Heuristics for vertex coloring include Brèlaz [Brè79], Matula [MMI72], and Turner [Tur88]. Wilf [Wil84] proved that backtracking to test whether a random graph has chromatic number k runs in constant time, dependent on k but independent of n. This is not as interesting as it sounds, because only a vanishingly small fraction of such graphs are indeed k-colorable. Expositions on algorithms to recognize bipartite graphs include [Man89]. Expositions on the hardness of 3-coloring graphs include [AHU74, Eve79a, Man89]. An interesting application of vertex coloring to scheduling traffic lights appears in [AHU83]. Baase [Baa88] gives a very good description of approximation algorithms for graph coloring, including approximation algorithm, where is the chromatic Wigderson's [Wig83] factor of number of G. Hardness of approximation results for vertex coloring include [BGS95]. Brook's theorem states that the chromatic number , where is the maximum degree of a vertex of G. Equality holds only for odd-length cycles (which have chromatic number 2) and complete graphs. The most famous problem in the history of graph theory is the four-color problem, first posed in 1852 and finally settled in 1976 by Appel and Haken using a proof involving extensive computation. Any planar graph can be 5-colored using a variation of the color interchange heuristic. Despite the four-color theorem, it is NP-complete to test whether a particular planar graph requires four colors or whether three suffice. See [SK86] for an exposition on the history of the four-color problem and the proof. An efficient algorithm to four-color a graph is presented in [RSST96]. Related Problems: Independent set (see page

), edge coloring (see page

).

Next: Edge Coloring Up: Graph Problems: Hard Problems Previous: Graph Partition Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE178.HTM (4 of 4) [19/1/2003 1:31:26]

Edge Coloring

Next: Graph Isomorphism Up: Graph Problems: Hard Problems Previous: Vertex Coloring

Edge Coloring

Input description: A graph G=(V,E). Problem description: What is the smallest set of colors needed to color the edges of E such that no two same-color edges share a vertex in common? Discussion: The edge coloring of graphs arises in a variety of scheduling applications, typically associated with minimizing the number of noninterfering rounds needed to complete a given set of tasks. For example, consider a situation where we need to schedule a given set of two-person interviews, where each interview takes one hour. All meetings could be scheduled to occur at distinct times to avoid conflicts, but it is less wasteful to schedule nonconflicting events simultaneously. We can construct a graph whose vertices are the people and whose edges represent the pairs of people who want to meet. An edge coloring of this graph defines the schedule. The color classes represent the different time periods in the schedule, with all meetings of the same color happening simultaneously. The National Football League solves such an edge coloring problem each season to make up its schedule. Each team's opponents are determined by the records of the previous season. Assigning the opponents to weeks of the season is the edge-coloring problem, presumably complicated by the constraints of spacing out rematches and making sure that there is a good game every Monday night.

file:///E|/BOOK/BOOK4/NODE179.HTM (1 of 3) [19/1/2003 1:31:27]

Edge Coloring

The minimum number of colors needed to edge color a graph is called by some its edge-chromatic number and others its chromatic index. To gain insight into edge coloring, note that a graph consisting of an even-length cycle can be edge-colored with 2 colors, while odd-length cycles have an edgechromatic number of 3. Edge coloring has a better (if less famous) theorem associated with it than does vertex coloring. Vizing's theorem states that any graph with a maximum vertex degree of can be edge colored using at most colors. To put this in perspective, note that any edge coloring must have at least colors, since each of the edges incident on any vertex must be distinct colors. Further, the proof of Vizing's theorem is constructive and can be turned into an algorithm to find an colors, which gives us an edge coloring using at most one extra color. Since it edge-coloring with is NP-complete to decide whether we can save this one color, it hardly seems worth the effort to worry about it. An implementation of Vizing's theorem is described below. Any edge coloring problem on G can be converted to the problem of finding a vertex coloring on the line graph L(G), which has a vertex of L(G) for each edge of G and an edge of L(G) if and only if the two edges of G share a common vertex. Line graphs can be constructed in time linear to their size, and any vertex coloring code from Section

can be employed to color them.

Although any edge coloring problem can be so formulated as a vertex coloring problem, this is usually a bad idea, since the edge coloring problem is easier to solve. Vizing's theorem is our reward for the extra thought needed to see that we have an edge coloring problem. Implementations: Yan Dong produced an implementation of Vizing's theorem in C++ as a course project for my algorithms course while a student at Stony Brook. It can be found on the algorithm repository WWW site http://www.cs.sunysb.edu/ algorith, as can an alternative program by Mark Goldberg and Amir Sehic. Combinatorica [Ski90] provides Mathematica implementations of edge coloring, via the line graph transformation and vertex coloring routines.

See Section

for further information on Combinatorica.

Notes: Graph-theoretic results on edge coloring are surveyed in [FW77]. Vizing [Viz64] and Gupta colors. Despite these tight [Gup66] both proved that any graph can be edge colored using at most bounds, Holyer [Hol81] proved that computing the edge-chromatic number is NP-complete. , any two Whitney, in introducing line graphs [Whi32], showed that with the exception of and connected graphs with isomorphic line graphs are isomorphic. It is an interesting exercise to show that the line graph of an Eulerian graph is both Eulerian and Hamiltonian, while the line graph of a Hamiltonian graph is always Hamiltonian. file:///E|/BOOK/BOOK4/NODE179.HTM (2 of 3) [19/1/2003 1:31:27]

Edge Coloring

Related Problems: Vertex coloring (see page

), scheduling (see page

).

Next: Graph Isomorphism Up: Graph Problems: Hard Problems Previous: Vertex Coloring Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE179.HTM (3 of 3) [19/1/2003 1:31:27]

Graph Isomorphism

Next: Steiner Tree Up: Graph Problems: Hard Problems Previous: Edge Coloring

Graph Isomorphism

Input description: Two graphs, G and H. Problem description: Find a (or all) mappings f of the vertices of G to the vertices of H such that G and H are identical; i.e. (x,y) is an edge of G iff (f(x),f(y)) is an edge of H. Discussion: Isomorphism is the problem of testing whether two graphs are really the same. Suppose we are given a collection of graphs and must perform some operation on each of them. If we can identify which of the graphs are duplicates, they can be discarded so as to avoid redundant work. We need some terminology to settle what is meant when we say two graphs are the same. Two labeled graphs and are identical when iff . The isomorphism problem consists of finding a mapping from the vertices of G to H such that they are identical. Such a mapping is called an isomorphism. Identifying symmetries is another important application of graph isomorphism. A mapping of a graph to itself is called an automorphism, and the collection of automorphisms (its automorphism group) provides a great deal of information about symmetries in the graph. For example, the complete graph has n! automorphisms (any mapping will do), while an arbitrary random graph is likely to have few or file:///E|/BOOK/BOOK4/NODE180.HTM (1 of 4) [19/1/2003 1:31:29]

Graph Isomorphism

perhaps only one, since G is always identical to itself. Several variations of graph isomorphism arise in practice: ●

Is graph G contained in (not identical to) graph H? - Instead of testing equality, we are often interested in knowing whether a small pattern graph G is a subgraph of H. Such problems as clique, independent set, and Hamiltonian cycle are important special cases of subgraph isomorphism. There are two distinct notions of ``contained in'' with respect to graphs. Subgraph isomorphism asks whether there is a subset of edges and vertices of G that is isomorphic to a smaller graph H. Induced subgraph isomorphism asks whether there is a subset of vertices of G whose deletion leaves a subgraph isomorphic to a smaller graph H. For induced subgraph isomorphism, all edges of G must be present in H, but also all nonedges of G must be nonedges of H. Clique happens to be an instance of both subgraph problems, while Hamiltonian cycle is an example of vanilla subgraph isomorphism.

●

●

Be aware of this distinction in your application. Subgraph isomorphism problems tend to be harder than graph isomorphism, and induced subgraph problems tend to be even harder than subgraph isomorphism. Backtracking is your only viable approach. Are your graphs labeled or unlabeled? - In many applications, vertices or edges of the graphs are labeled with some attribute that must be respected in determining isomorphisms. For example, in comparing two bipartite graphs, each with ``worker'' vertices and ``job'' vertices, any isomorphism that equated a job with a worker would make no sense. Labels and related constraints can be factored into any backtracking algorithm. Further, such constraints can be used to significantly speed up the search by creating more opportunities for pruning whenever two vertex labels do not match up. Are you testing whether two trees are isomorphic? - There are faster algorithms for certain special cases of graph isomorphism, such as trees and planar graphs. Perhaps the most important case is detecting isomorphisms among trees, a problem that arises in language pattern matching and parsing applications. A parse tree is often used to describe the structure of a text; two parse trees will be isomorphic if the underlying pair of texts have the same structure. Efficient algorithms for tree isomorphism begin with the leaves of both trees and work inward towards the center. Each vertex in one tree is assigned a label representing the set of vertices in the second tree that might possibly be isomorphic to it, based on the constraints of labels and vertex degrees. For example, all the leaves in tree are initially potentially equivalent to all leaves of . Now, working inward, we can partition the vertices adjacent to leaves in into classes based on how many leaves and nonleaves they are adjacent to. By carefully keeping track of the labels of the subtrees, we can make sure that we have the same distribution of labeled , while completing the process partitions the subtrees for and . Any mismatch means

file:///E|/BOOK/BOOK4/NODE180.HTM (2 of 4) [19/1/2003 1:31:29]

Graph Isomorphism

vertices into equivalence classes defining all isomorphisms. See the references below for more details. No polynomial-time algorithm is known for graph isomorphism, but neither is it known to be NP), it one of the few important algorithmic complete. Along with integer factorization (see Section problems whose rough computational complexity is still not known. The conventional wisdom is that isomorphism is a problem that lies between P and NP-complete if P NP. Although no worst-case polynomial-time algorithm is known, testing isomorphism in practice is usually not very hard. The basic algorithm backtracks through all n! possible relabelings of the vertices of graph h with the names of vertices of graph g, and then tests whether the graphs are identical. Of course, we can prune the search of all permutations with a given prefix as soon as we detect any mismatch between edges both of whose vertices are in the prefix. However, the real key to efficient isomorphism testing is to preprocess the vertices into ``equivalence classes'', partitioning them into sets of vertices such that two vertices in different sets cannot possibly be mistaken for each other. All vertices in each equivalence class must share the same value of some invariant that is independent of labeling. Possibilities include: ●

●

●

Vertex degree - This simplest way to partition vertices is based on their degree, the number of edges incident on the vertex. Clearly, two vertices of different degree cannot be identical. This simple partition can often be a big win, but it won't do much for regular graphs, where each vertex has the same degree. Shortest path matrix - For each vertex v, the all-pairs shortest path matrix (see Section ) defines a multiset of n-1 distances (possibly with repeats) representing the distances between v and each of the other vertices. Any two vertices that are identical in isomorphic graphs will define the exact same multiset of distances, so we can partition the vertices into equivalence classes defining identical distance multisets. Counting length-k paths - Taking the adjacency matrix of G and raising it to the kth power gives a matrix where counts the number of (nonsimple) paths from i to j. For each vertex and each k, this matrix defines a multiset of path-counts, which can be used for partitioning as with or beyond, and use any single deviation as an excuse distances above. You could try all to partition.

Using these invariants, you should be able to partition the vertices of each graph into a large number of small equivalence classes. Finishing the job off with backtracking, using the name of each equivalence class as a label, should usually be quick work. If the sizes of the equivalence classes of both graphs are not identical, then the graphs cannot be isomorphic. It is harder to detect isomorphisms between graphs with high degrees of symmetry than it is for arbitrary graphs, because of the effectiveness of these equivalence-class partitioning heuristics.

file:///E|/BOOK/BOOK4/NODE180.HTM (3 of 4) [19/1/2003 1:31:29]

Graph Isomorphism

Implementations: The world's fastest isomorphism testing program is Nauty, by Brendan D. McKay. Nauty (No AUTomorphisms, Yes?) is a set of very efficient C language procedures for determining the automorphism group of a vertex-colored graph. Nauty is also able to produce a canonically labeled isomorph of the graph, to assist in isomorphism testing. It was the basis of the first program to generate all the 11-vertex graphs without isomorphs, and can test most graphs of fewer than one hundred vertices in well under a second. Nauty has been successfully ported to a variety of operating systems and C compilers. It may be obtained from http://cs.anu.edu.au/people/bdm/. It is free for educational and research applications, but for commercial use contact the author at [email protected]. Combinatorica [Ski90] provides (slow) Mathematica implementations of graph isomorphism and automorphism testing.

See Section

for further information on Combinatorica.

Notes: Graph isomorphism is an important problem in complexity theory. Monographs on isomorphism detection include Hoffmann [Hof82]. Polynomial-time algorithms are known for planar graph isomorphism [HW74] and for graphs where the maximum vertex degree is bounded by a constant [Luk80]. The all-pairs shortest path heuristic is due to [SD76], although there exist nonisomorphic graphs that realize the same set of distances [BH90]. A linear-time tree isomorphism algorithm for both labeled and unlabeled trees is presented in [AHU74]. A problem is said to be isomorphism-complete if it is provably as hard as isomorphism. Testing the isomorphism of bipartite graphs is isomorphism-complete, since any graph can be made bipartite by replacing each edge by two edges connected with a new vertex. Clearly, the original graphs are isomorphic if and only if the transformed graphs are. Related Problems: Shortest path (see page

), string matching (see page

).

Next: Steiner Tree Up: Graph Problems: Hard Problems Previous: Edge Coloring Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE180.HTM (4 of 4) [19/1/2003 1:31:29]

Steiner Tree

Next: Feedback Edge/Vertex Set Up: Graph Problems: Hard Problems Previous: Graph Isomorphism

Steiner Tree

Input description: A graph G=(V,E). A subset of vertices

.

Problem description: Find the smallest tree connecting all the vertices of T. Discussion: Steiner tree often arises in network design and wiring layout problems. Suppose we are given a set of sites that must be connected by wires as cheaply as possible. The minimum Steiner tree describes the way to connect them using the smallest amount of wire. Analogous problems arise in designing networks of water pipes or heating ducts in buildings. Similar considerations also arise in VLSI circuit layout, where we seek to connect a set of sites to (say) ground under constraints such as material cost, signal propagation time, or reducing capacitance. The Steiner tree problem is distinguished from the minimum spanning tree problem (see Section ) in that we are permitted to construct or select intermediate connection points to reduce the cost of the tree. Issues in Steiner tree construction include: ●

How many points do you have to connect? - The Steiner tree of a pair of vertices is simply the shortest path between them (see Section

file:///E|/BOOK/BOOK4/NODE181.HTM (1 of 4) [19/1/2003 1:31:31]

). The Steiner tree of all the vertices, when S=V,

Steiner Tree

●

●

●

●

simply defines the minimum spanning tree of G. Despite these special cases, the general minimum Steiner tree problem is NP-hard and remains so under a broad range of restrictions. Is the input a set of geometric points or a distance graph? - Geometric versions of Steiner tree take as input a set of points, typically in the plane, and seek the smallest tree connecting the points. A complication is that the set of possible intermediate points is not given as part of the input but must be deduced from the set of points. These possible Steiner points must satisfy several geometric properties, which can be used to reduce the set of candidates down to a finite number. For example, every Steiner point will have degree exactly three in a minimum Steiner tree, and the angles formed between any two of these edges will be exactly 120 degrees. Are there constraints on the edges we can use? - Many wiring problems correspond to geometric versions of the problem where all edges are restricted to being either horizontal or vertical, which is the so-called rectilinear Steiner problem. A different set of angular and degree conditions applies for rectilinear Steiner trees than for Euclidean trees. In particular, all angles must be multiples of 90 degrees, and each vertex is of degree up to four. Do I really need an optimal tree? - In certain applications, such as minimum cost communications networks, construction costs are high enough to invest large amounts of computation in finding the best possible Steiner tree. This implies an exhaustive search technique such as backtracking or branch-and-bound. There are many opportunities for pruning search based on geometric constraints. For graph instances, network reduction procedures can reduce the problem to a graph typically one-quarter the size of the input graph. Still, Steiner tree remains a hard problem. Through exhaustive search methods, instances as large as 32 points for the Euclidean and 30 for the rectilinear problems can be confidently solved to optimality. We recommend experimenting with the implementations described below before attempting your own. How can I reconstruct Steiner vertices I never knew about? - A very special type of Steiner tree arises in classification and evolution. A phylogenic tree illustrates the relative similarity between different objects or species. Each object represents (typically) a terminal vertex of the tree, with intermediate vertices representing branching points between classes of objects. For example, an evolutionary tree of animal species might have leaf nodes including (human, dog, snake) and internal nodes corresponding to the taxa (animal, mammal, reptile). A tree rooted at animal with dog and human classified under mammal implies that humans are closer to dogs than to snakes. Many different phylogenic tree construction algorithms have been developed, which vary in the data they attempt to model and what the desired optimization criterion is. Because they all give different answers, identifying the correct algorithm for a given application is somewhat a matter of faith. A reasonable procedure is to acquire a standard package of implementations, discussed below, and then see what happens to your data under all of them.

Fortunately, there is a good, efficient heuristic for finding Steiner trees that works well on all versions of the problem. Construct a graph modeling your input, with the weight of edge (i,j) equal to the distance from point i to point j. Find a minimum spanning tree of this graph. You are guaranteed a provably good approximation for both Euclidean and rectilinear Steiner trees. file:///E|/BOOK/BOOK4/NODE181.HTM (2 of 4) [19/1/2003 1:31:31]

Steiner Tree

The worst case for a minimum spanning tree approximation of the Euclidean distance problem is three points forming an equilateral triangle. The minimum spanning tree will contain two of the sides (for a length of 2), whereas the minimum Steiner tree will connect the three points using an interior point, for a total length of . This ratio of is always achieved, and in practice the easily-computed minimum spanning tree is usually within a few percent of the optimal Steiner tree. For rectilinear Steiner trees, the ratio with rectilinear minimum spanning trees is always . Such a minimum spanning tree can be refined by inserting a Steiner point whenever the edges of the minimum spanning tree incident on a vertex form an angle of less than 120 degrees between them. Inserting these points and locally readjusting the tree edges can move the solution a few more percent towards the optimum. Similar optimizations are possible for rectilinear spanning trees. An alternative heuristic for graphs is based on shortest path. Start with a tree consisting of the shortest path between two terminals. For each remaining terminal t, find the shortest path to a vertex v in the tree and add this path to the tree. The time complexity and quality of this heuristic depend upon the insertion order of the terminals and how the shortest-path computations are performed, but something simple and fairly effective is likely to result. Implementations: Salowe and Warme [SW95] have developed a program for computing exact rectilinear Steiner minimal trees. It is available by anonymous ftp from ftp.cs.virginia.edu in the directory pub/french/salowe/newsteiner.tar.Z. It should be capable of handling up to 30 points routinely. A heuristic program by Robins and Zhang is available from the algorithm repository http://www.cs.sunysb.edu/ algorith. PHYLIP is an extensive and widely used package of programs for inferring phylogenic trees. It contains over twenty different algorithms for constructing phylogenic trees from data. Although many of them are designed to work with molecular sequence data, several general methods accept arbitrary distance matrices as input. With versions written in C and Pascal, it is available on the WWW from http://evolution.genetics.washington.edu/phylip.html. Notes: The most complete reference on the Steiner tree problem is the monograph by Hwang, Richards, and Winter [HRW92]. Surveys on the problem include [Kuh75]. Steiner tree problems arising in VLSI design are discussed in [KPS89, Len90]. Empirical results on Steiner tree heuristics include [SFG82, Vos92]. The Euclidean Steiner problem dates back to Fermat, who asked how to find a point p in the plane minimizing the sum of the distances to three given points. This was solved by Torricelli before 1640. Steiner was apparently one of several mathematicians who worked the general problem for n points, and he was mistakenly credited with the problem. An interesting, more detailed history appears in [HRW92].

file:///E|/BOOK/BOOK4/NODE181.HTM (3 of 4) [19/1/2003 1:31:31]

Steiner Tree

Gilbert and Pollak [GP68] first conjectured that the ratio of the length of the minimum Steiner tree over . After twenty years of active research, the Gilbertthe minimum spanning tree is always Pollak ratio was finally proven by Du and Hwang [DH92]. The Euclidean minimum spanning tree for n time [PS85]. points in the plane can be constructed in Expositions on the proof that the Steiner tree problem for graphs is hard [Kar72] include [Eve79a]. Expositions on exact algorithms for Steiner trees in graphs include [Law76]. The hardness of Steiner tree for Euclidean and rectilinear metrics was established in [GGJ77, GJ77]. Euclidean Steiner tree is not known to be in NP, because of numerical issues in representing distances. Analogies can be drawn between minimum Steiner trees and minimum energy configurations in certain physical systems. The case that such analog systems, including the behavior of soap films over wire frames, ``solve'' the Steiner tree problem is discussed in [Mie58]. Related Problems: Minimum spanning tree (see page

), shortest path (see page

).

Next: Feedback Edge/Vertex Set Up: Graph Problems: Hard Problems Previous: Graph Isomorphism Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE181.HTM (4 of 4) [19/1/2003 1:31:31]

Feedback Edge/Vertex Set

Next: Computational Geometry Up: Graph Problems: Hard Problems Previous: Steiner Tree

Feedback Edge/Vertex Set

Input description: A (directed) graph G=(V,E). Problem description: What is the smallest set of edges E' or vertices V' whose deletion leaves an acyclic graph? Discussion: Feedback set problems arise because many algorithmic problems are much easier or much better defined on directed acyclic graphs than on arbitrary digraphs. Topological sorting (see Section ) can be used to test whether a graph is a DAG, and if so, to order the vertices so as to respect the edges as precedence scheduling constraints. But how can you design a schedule if there are cyclic constraints, such as A must be done before B, which must be done before C, which must be done before A? By identifying a feedback set, we identify the smallest number of constraints that must be dropped so as to permit a valid schedule. In the feedback edge (or arc) set problem, we drop precedence constraints (job A must come before job B). In the feedback vertex set problem, we drop entire jobs and any associated constraints. It is also referred to in the literature as the maximum acyclic subgraph problem. ●

Do any constraints have to be dropped? - Not if the graph is a DAG, which can be tested via

file:///E|/BOOK/BOOK4/NODE182.HTM (1 of 3) [19/1/2003 1:31:32]

Feedback Edge/Vertex Set

●

●

●

topological sort in linear time, Further, topological sorting also gives a simple way to find a feedback set if we modify the algorithm to delete the edge or vertex whenever a contradiction is found instead of simply printing a warning. The catch is that this feedback set might be much larger than needed, no surprise since both feedback edge set and feedback vertex set are NPcomplete on directed graphs. How can I find a good feedback edge set? - A simple but effective linear-time heuristic constructs a vertex ordering, just as in the topological sort heuristic above, and deletes any arc going from right to left. This heuristic builds up the ordering from the outside in based on the in- and outdegrees of each vertex. Any vertex of in-degree 0 is a source and can be placed first in the ordering. Any vertex of out-degree 0 is a sink and can be placed last in the ordering, again without violating any constraints. If not, we find the vertex with the maximum difference between in- and out-degree, and place it on the side of the permutation that will salvage the greatest number of constraints. Delete any vertex from the DAG after positioning it and repeat until the graph is empty. How can I find a good feedback vertex set? - The following variant of the above heuristic should be effective. Keep any source or sink we encounter. If none exist, add to the feedback set a vertex v that maximizes , since this vertex is furthest from becoming either a source or sink. Again, delete any vertex from the DAG after positioning it and repeat until the graph is empty. What if I want to break all cycles in an undirected graph? - The problem of finding feedback sets from undirected graphs is different for digraphs, and in one case actually easier. An undirected graph without cycles is a tree. It is well known that any tree on n vertices has exactly n-1 edges. Thus the smallest feedback edge set of any undirected graph is |E| - (n-1), and it can be found by deleting all the edges not in any given spanning tree. Spanning trees can be most efficiently constructed using depth-first search, as discussed in Section remains NP-complete for undirected graphs, however.

. The feedback vertex set problem

Finding an optimal feedback set is NP-complete, and for most applications it is unlikely to be worth searching for the smallest set. However, in certain applications it would pay to try to refine the heuristic solutions above via randomization or simulated annealing. To move between states, we modify the vertex permutation by swapping pairs in order or inserting/deleting vertices into the feedback set. Implementations: The econ_order program of the Stanford GraphBase (see Section ) permutes the rows and columns of a matrix so as to minimize the sum of the numbers below the main diagonal. Using an adjacency matrix as the input and deleting all edges below the main diagonal leaves an acyclic graph. Notes: The feedback set problem first arose in [Sla61]. Heuristics for feedback set problems include [BYGNR94, ELS93, Fuj96]. Expositions of the proofs that feedback minimization is hard [Kar72] include [AHU74, Eve79a]. Both feedback vertex and edge set remain hard even if no vertex has indegree or out-degree greater than two [GJ79].

file:///E|/BOOK/BOOK4/NODE182.HTM (2 of 3) [19/1/2003 1:31:32]

Feedback Edge/Vertex Set

An interesting application of feedback arc set to economics is presented in [Knu94]. For each pair A,B of sectors of the economy, we are given how much money flows from A to B. We seek to order the sectors to determine which sectors are primarily producers to other sectors, and which deliver primarily to consumers. Related Problems: Bandwidth reduction (see page (see page

), topological sorting (see page

), scheduling

).

Next: Computational Geometry Up: Graph Problems: Hard Problems Previous: Steiner Tree Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE182.HTM (3 of 3) [19/1/2003 1:31:32]

Computational Geometry

Next: Robust Geometric Primitives Up: A Catalog of Algorithmic Previous: Feedback Edge/Vertex Set

Computational Geometry Computational geometry is the algorithmic study of geometric problems and objects. Compared to the other topics in this book, computational geometry emerged as a field quite recently, with Shamos's Ph.D. thesis [Sha78] typically cited as its founding event. Its emergence coincided with the explosion of computer graphics and windowing systems, which directly or indirectly provide much of the motivation for geometric computing. The past twenty years have seen enormous growth in computational geometry, resulting in a significant body of useful algorithms, software, textbooks, and research results. Good books on computational geometry include: ●

●

●

●

●

Preparata and Shamos [PS85] - Although aging a bit, this book remains the best general introduction to computational geometry, stressing algorithms for convex hulls, Voronoi diagrams, and intersection detection. O'Rourke [O'R94] - Perhaps the best practical introduction to computational geometry. The emphasis is on careful and correct implementation (in C language) of the fundamental algorithms of computational geometry. These implementations are available from http://grendel.csc.smith.edu/ orourke/. Edelsbrunner [Ede87] - This is the definitive book on arrangements, a topic that runs through most of computational geometry. Although not appropriate for beginners, it provides an important perspective for advanced geometers. Mulmuley [Mul94] - An approach to computational geometry through randomized incremental algorithms. Very interesting, but likely too narrow to serve as a general introduction. Nievergelt and Hindrichs [NH93] - This idiosyncratic algorithms text focuses on problems in graphics and geometry. Good coverage of line drawing, intersection algorithms, and spatial data structures, but with too many topics touched on too lightly to serve as an effective reference.

The leading conference in computational geometry is the ACM Symposium on Computational Geometry, held annually in late May or early June. Although the primary results presented at the conference are theoretical, there has been a concerted effort on the part of the research community to increase the presence of applied, experimental work through video reviews and poster sessions. The other major annual conference is the Canadian Conference on Computational Geometry (CCCG), typically held in early August. Useful literature surveys include [Yao90]. A unique source of computational geometry information is geom.bib, a community effort to maintain a file:///E|/BOOK/BOOK4/NODE183.HTM (1 of 3) [19/1/2003 1:31:33]

Computational Geometry

complete bibliography on computational geometry. It references over eight thousand books, papers, and reports and includes detailed abstracts for many of them. Grep-ing through the geom.bib is an amazingly efficient way to find out about previous work without leaving your office. It is available via anonymous ftp from ftp.cs.usask.ca, in the file pub/geometry/geombib.tar.Z. There is a growing body of implementations of geometric algorithms. We point out specific implementations where applicable in the catalog, but the reader should be aware of three specific WWW sites: ●

●

●

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

The Geometry Center's directory of computational geometry software, maintained by Nina Amenta, is the ``official'' site for all computational geometry software. Check here first to see what is available: http://www.geom.umn.edu/software/cglist/. Graphics Gems is a series of books dedicated to collecting small codes of interest in computer graphics. Many of these programs are geometric in nature. All associated codes are available from ftp://ftp-graphics.stanford.edu/pub/Graphics/GraphicsGems. CGAL (Computational Geometry Algorithms Library) is a joint European project now underway to produce a comprehensive library of geometric algorithms. This will likely become the definitive geometric software project. Check out its progress at http://www.cs.ruu.nl/CGAL/.

Robust Geometric Primitives Convex Hull Triangulation Voronoi Diagrams Nearest Neighbor Search Range Search Point Location Intersection Detection Bin Packing Medial-Axis Transformation Polygon Partitioning Simplifying Polygons Shape Similarity Motion Planning Maintaining Line Arrangements Minkowski Sum

file:///E|/BOOK/BOOK4/NODE183.HTM (2 of 3) [19/1/2003 1:31:33]

Computational Geometry

Next: Robust Geometric Primitives Up: A Catalog of Algorithmic Previous: Feedback Edge/Vertex Set Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE183.HTM (3 of 3) [19/1/2003 1:31:33]

Robust Geometric Primitives

Next: Convex Hull Up: Computational Geometry Previous: Computational Geometry

Robust Geometric Primitives

Input description: A point p and a line segment l, or two line segments Problem description: Does p lie over, under, or on l? Does

intersect

and

.

?

Discussion: Implementing basic geometric primitives is a task fraught with peril, even for such simple tasks as returning the intersection point of two lines. What should you return if the two lines are parallel, meaning they don't intersect at all? What if the lines are identical, so the intersection is not a point but the entire line? What if one of the lines is horizontal, so that in the course of solving the equations for the intersection point you are likely to divide by zero? What if the two lines are almost parallel, so that the intersection point is so far from the origin as to cause arithmetic overflows? These issues become even more complicated for intersecting line segments, since there are a bunch of other special cases that must be watched for and treated specially. If you are new to implementing geometric algorithms, I suggest that you study O'Rourke's Computational Geometry in C [O'R94] for practical advice and complete implementations of basic geometric algorithms and data structures. You are likely to avoid many headaches by following in his footsteps.

file:///E|/BOOK/BOOK4/NODE184.HTM (1 of 5) [19/1/2003 1:31:35]

Robust Geometric Primitives

There are two different issues at work here: geometric degeneracy and numerical stability. Degeneracy refers to annoying special cases that must be treated in substantially different ways, such as when two lines intersect in more or less than a single point. There are three primary approaches to dealing with degeneracy: ●

●

●

Ignore it - Make as an operating assumption that your program will work correctly only if no three points are collinear, no three lines meet at a point, no intersections happen at the endpoints of line segments, etc. This is probably the most common approach, and what I would recommend for short-term projects if you can live with frequent crashes. The drawback is that interesting data often comes from points sampled on a grid and is inherently very degenerate. Fake it - Randomly or symbolically perturb your data so that it seems nondegenerate. By moving each of your points a small amount in a random direction, you can break many of the existing degeneracies in the data, hopefully without creating too many new problems. This probably should be the first thing to try as soon as you decide that your program is crashing too often. One problem with random perturbations is that they do change the shape of your data in subtle ways, which may be intolerable for your application. There also exist techniques to ``symbolically'' perturb your data to remove degeneracies in a consistent manner, but these require serious study to apply correctly. Deal with it - Geometric applications can be made more robust by writing special code to handle each of the special cases that arise. This can work well if done with care at the beginning, but not so well if kludges are added whenever the system crashes. Expect to expend a lot of effort if you are determined to do it right.

Geometric computations often involve floating-point arithmetic, which leads to problems with overflows and numerical precision. There are three basic approaches to the issue of numerical stability: ●

●

●

Integer arithmetic - By forcing all points of interest to lie on a fixed-size integer grid, you can perform exact comparisons to test whether any two points are equal or two line segments intersect. The cost is that the intersection point of two lines may not be exactly representable. This is likely to be the simplest and best method, if you can get away with it. Double precision reals - By using double-precision floating point numbers, you may get lucky and avoid numerical errors. Your best bet might be to keep all the data as single-precision reals, and use double-precision for intermediate computations. Arbitrary precision arithmetic - This is certain to be correct, but also to be slow. This approach seems to be gaining favor in the research community with the observation that careful analysis can minimize the need for high-precision arithmetic, and thus the performance penalty. Still, you should expect high-precision arithmetic to be several orders of magnitude slower than standard floating-point arithmetic.

The difficulties associated with producing robust geometric software are still under attack by researchers. The best practical technique is to base your applications on a small set of geometric primitives that handle as much of the low-level geometry as possible. These primitives include:

file:///E|/BOOK/BOOK4/NODE184.HTM (2 of 5) [19/1/2003 1:31:35]

Robust Geometric Primitives ●

Area of a triangle - Although it is well-known that the area A(t) of a triangle t=(a,b,c) is half the base times the height, computing the length of the base and altitude is messy work with trigonometric functions. It is better to use the determinant formula for twice the area:

This formula generalizes to compute d! times the volume of a simplex in d dimensions. Thus 3! = 6 times the volume of a tetrahedron t=(a,b,c,d) in three dimensions is

Note that these are signed volumes and can be negative, so take the absolute value first. Section explains how to compute determinants.

●

●

●

The conceptually simplest way to compute the area of a polygon (or polyhedron) is to triangulate it and then sum up the area of each triangle. An implementation of a slicker algorithm that avoids triangulation is discussed in [O'R94]. Above-below-on test - Does a given point c lie above, below, or on a given line l? A clean way to deal with this is to represent l as a directed line that passes through point a before point b, and ask whether c lies to the left or right of the directed line l. It is up to you to decide whether left means above or below. This primitive can be implemented using the sign of the area of a triangle as computed above. If the area of t(a,b,c) > 0, then c lies to the left of . If the area of t(a,b,c) = 0, then c lies on . Finally, if the area of t(a,b,c) < 0, then c lies to the right of . This generalizes naturally to three dimensions, where the sign of the area denotes whether d lies above or below the oriented plane (a,b,c). Line segment intersection - The above-below primitive can be used to test whether a line intersects a line segment. It does iff one endpoint of the segment is to the left of the line and the other is to the right. Segment intersection is similar but more complicated, and we refer you to implementations described below. The decision whether two segments intersect if they share an endpoint depends upon your application and is representative of the problems of degeneracy. In-circle test - Does the point d lie inside or outside the circle defined by points a, b, and c in the plane? This primitive occurs in all Delaunay triangulation algorithms and can be used as a robust way to do distance comparisons. Assuming that a, b, c are labeled in counterclockwise order around the circle, compute the determinant:

file:///E|/BOOK/BOOK4/NODE184.HTM (3 of 5) [19/1/2003 1:31:35]

Robust Geometric Primitives

Incircle will return 0 if all four points are cocircular, a positive value if d is inside the circle, and negative if d is outside. Check out the implementations described below before you attempt to build your own. Implementations: LEDA (see Section ) provides a very complete set of geometric primitives for planar geometry, written in C++. If you are writing a significant geometric application, you should consider basing it on LEDA. At least check them out before you try to write your own. O'Rourke [O'R94] provides implementations in C of most of the primitives discussed in this section. See Section . These primitives were implemented primarily for exposition rather than production use, but they should be quite reliable and might be more appropriate than LEDA for small applications. A robust implementation of the basic geometric primitives in C++ using exact arithmetic, by Jonathan Shewchuk, is available at http://www.cs.cmu.edu/ quake/robust.html. Don't expect them to be very fast. Pascal implementations of basic geometric primitives appear in [MS91]. Sedgewick [Sed92] provides fragments of the basic primitives in C++. See Section

for both of them.

An alternative C++ library of geometric algorithms and data structures (although you are almost certainly better off sticking to LEDA) is Geolab, written by Pedro J. de Rezende, Welson R. Jacometti, Cesar N. Gon, and Laerte F. Morgado, Universidade Estadual de Campinas, Brazil. Geolab requires the SUN C++ compiler, but a Sparc binary and visualization environment is included along with all source code. Geolab appears to be primarily for the brave, since its robustness is uncertain and it contains little documentation, but it does provide 40 algorithms, including such advanced topics as farthest point Voronoi diagrams, nearest neighbor search, and ray shooting. Notes: O'Rourke [O'R94] provides an implementation-oriented introduction to computational geometry, which stresses robust geometric primitives and is recommended reading. Shewchuk [She96] and Fortune and van Wyk [FvW93] present careful studies on the costs of using arbitrary-precision arithmetic for geometric computation. By being careful about when to use it, reasonable efficiency can be maintained while achieving complete robustness. Other approaches to achieving robustness include [DS88, Hof89, Mil89]. file:///E|/BOOK/BOOK4/NODE184.HTM (4 of 5) [19/1/2003 1:31:35]

Robust Geometric Primitives

Related Problems: Intersection detection (see page

), maintaining arrangements (see page

Next: Convex Hull Up: Computational Geometry Previous: Computational Geometry Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE184.HTM (5 of 5) [19/1/2003 1:31:35]

).

Convex Hull

Next: Triangulation Up: Computational Geometry Previous: Robust Geometric Primitives

Convex Hull

Input description: A set S of n points in d-dimensional space. Problem description: Find the smallest convex polygon containing all the points of S. Discussion: Finding the convex hull of a set of points is the most elementary interesting problem in computational geometry, just as minimum spanning tree is the most elementary interesting problem in graph algorithms. It arises because the hull quickly captures a rough idea of the shape or extent of a data set. Convex hull also serves as a first preprocessing step to many, if not most, geometric algorithms. For example, consider the problem of finding the diameter of a set of points, which is the pair of points a maximum distance apart. The diameter will always be the distance between two points on the convex hull. The algorithm for computing diameter proceeds by first constructing the convex hull, then for each hull vertex finding which other hull vertex is farthest away from it. This so-called ``rotatingcalipers'' method can be used to move efficiently from one hull vertex to another. There are almost as many convex hull algorithms as there are sorting algorithms. Answer the following questions to help choose between them: file:///E|/BOOK/BOOK4/NODE185.HTM (1 of 5) [19/1/2003 1:31:37]

Convex Hull

●

How many dimensions are you working with? - Convex hulls in two and even three dimensions are fairly easy to work with. However, as the dimension of a space increases, certain assumptions that were valid in lower dimensions break down. For example, any n-vertex polygon in two dimensions has exactly n edges. However, the relationship between the numbers of faces and vertices is more complicated even in three dimensions. A cube has 8 vertices and 6 faces, while an octahedron has 8 faces and 6 vertices. This has implications for data structures that represent hulls - are you just looking for the hull points or do you need the defining polyhedron? The need to find convex hulls in high-dimensional spaces arises in many applications, so be aware of such complications if your problem takes you there. Gift-wrapping is the basic algorithm for constructing higher-dimensional convex hulls. Observe that a three-dimensional convex polyhedron is composed of two-dimensional faces, or facets, which are connected by one-dimensional lines, or edges. Each edge joins exactly two facets together. Gift-wrapping starts by finding an initial facet associated with the lowest vertex and then conducting a breadth-first search from this facet to discover new, additional facets. Each edge e defining the boundary of a facet must be shared with one other facet, so by running through each of the n points we can identify which point defines the next facet with e. Essentially, we ``wrap'' the points one facet at a time by bending the wrapping paper around an edge until it hits the first point. The key to efficiency is making sure that each edge is explored only once. Implemented properly in d dimensions, gift-wrapping takes , where is the number of facets and is the number of edges in the convex hull. Thus gift-wrapping can be very efficient when there is only a constant number of facets on the hull. However, this can be as bad as when the convex hull is very complex.

●

●

Better convex hull algorithms are available for the important special case of three dimensions, where time in fact suffices. For three or higher dimensions, I recommend that you use one of the codes described below rather than roll your own. Is your data given as vertices or half-spaces? - The problem of finding the intersection of a set of n half-spaces in d dimensions is dual to that of computing convex hulls of n points in d dimensions. Thus the same basic algorithm suffices for both problems. The necessary duality . transformation is discussed in Section How many points are likely to be on the hull? - If your points are selected ``randomly'', it is likely that most of them lie within the interior of the hull. Planar convex hull programs can be made more efficient in practice using the observation than the leftmost, rightmost, topmost, and bottommost points must all be on the convex hull. Unless the topmost is leftmost and bottommost is rightmost, this gives a set of either three or four distinct points, defining a triangle or quadrilateral. Any point inside this region cannot be on the convex hull and can be discarded in a linear sweep through the points. Ideally, only a few points will then remain to run through the full convex hull algorithm.

file:///E|/BOOK/BOOK4/NODE185.HTM (2 of 5) [19/1/2003 1:31:37]

Convex Hull

●

This trick can also be applied beyond two dimensions, although it loses effectiveness as the dimension increases. How do I find the shape of my point set? - Although convex hulls provide a gross measure of shape, any details associated with concavities are lost. For example, the shape of the `G' would be indistinguishable from the shape of the `O'. A more general structure, called alpha-shapes, can be parameterized so as to retain arbitrarily large concavities. Implementations and references on alpha-shapes are included below.

The primary convex hull algorithm in the plane is the Graham scan. Graham scan starts with one point p known to be on the convex hull (say the point with lowest x-coordinate) and sorts the rest of the points in angular order around p. Starting with a hull consisting of p and the point with the smallest angle, we proceed counterclockwise around v adding points. If the angle formed by the new point and the last hull edge is less than 180 degrees, we add this new point to the hull. If the angle formed by the new point and the last ``hull'' edge is greater than 180 degrees, then a chain of vertices starting from the last hull edge must be deleted to maintain convexity. The total time is , since the bottleneck is sorting the points around v. The basic Graham scan procedure can also be used to construct a nonself-intersecting (or simple) polygon passing through all the points. Sort the points around v, but instead of testing angles simply connect the points in angular order. Connecting this to v gives a polygon without self-intersection, although it typically has many skinny protrusions. The gift-wrapping algorithm becomes especially simple in two dimensions, since each ``facet'' becomes an edge, each ``edge'' becomes a vertex of the polygon, and the ``breadth-first search'' simply walks around the hull in a clockwise or counterclockwise order. The 2D gift-wrapping, or Jarvis march, algorithm runs in O(n h) time, where h is the number of vertices on the convex hull. I would recommend sticking with Graham scan unless you really know in advance that there cannot be too many vertices on the hull. Implementations: O'Rourke [O'R94] provides a robust implementation of the Graham scan in two implementation of an incremental algorithm for convex hulls in three dimensions and an dimensions. Both are written in C. The latter has been proven capable of solving 10,000-point problems in a few minutes on a modern workstation. See Section

.

Qhull [BDH97] appears to be the convex hull code of choice for general dimensions (in particular from 2 to about 8 dimensions). It is written in C and can also construct Delaunay triangulations, Voronoi vertices, furthest-site Voronoi vertices, and half-space intersections. Qhull has been widely used in scientific applications and has a well-maintained home page at http://www.geom.umn.edu/software/qhull/. An alternative higher-dimensional convex hull code in ANSI C is Ken Clarkson's Hull, available at file:///E|/BOOK/BOOK4/NODE185.HTM (3 of 5) [19/1/2003 1:31:37]

Convex Hull

http://www.cs.att.com/netlib/voronoi/hull.html. It does not appear to be as widely used or actively maintained as Qhull, but it also does alpha-shapes. For an excellent alpha-shapes code, originating from the work of Edelsbrunner and Mucke [EM94], check out http://fiaker.ncsa.uiuc.edu/alpha/. Fukuda's cdd program is the best choice for nonsimplicial polytopes in about 6D and higher. See ftp://ifor13.ethz.ch/pub/fukuda/cdd/. It may be used for computing convex hulls and half-space intersection. XTango (see Section plane.

) provides animations of the Graham scan and Jarvis march algorithms in the

A Pascal implementation of Graham scan appears in [MS91]. See Section

. C++ implementations of

planar convex hulls includes LEDA (see Section ). Algorithm 523 [Edd77] of the Collected Algorithms of the ACM is a Fortran code for planar convex hulls. It is available from Netlib (see Section ). Notes: Constructing planar convex hulls plays a similar role in computational geometry as sorting does in algorithm theory. Like sorting, convex hull is a fundamental problem for which a wide variety of different algorithmic approaches lead to interesting or optimal algorithms. Preparata and Shamos [PS85] give a good exposition of several such algorithms, including quickhull and mergehull, both inspired by the sorting algorithms. In fact, a simple construction involving points on a parabola reduces sorting to convex hull, so the information-theoretic lower bound for sorting implies that planar convex hull requires time to compute. A stronger lower bound is established in [Yao81]. Good expositions of the Graham scan algorithm [Gra72] and the Jarvis march [Jar73] include [CLR90, PS85]. The gift wrapping algorithm was introduced by Chand and Kapur [CK70]. Noteworthy among time, where h is the planar convex hull algorithms is Seidel and Kirkpatrick [KS86], which takes number of hull vertices, which captures the best performance of both Graham scan and gift wrapping and is (theoretically) better in between. Alpha-hulls, introduced in [EKS83], provide a useful notion of the shape of a point set. A generalization to three dimensions, with an implementation, is presented in [EM94]. Reverse-search algorithms for constructing convex hulls are effective in higher dimensions [AF92], although constructions demonstrating the poor performance of convex hull algorithms for nonsimplicial polytopes are presented in [AB95]. Through a clever lifting-map construction [ES86], the problem of building Voronoi diagrams in d-dimensions can be reduced to constructing convex hulls in (d+1)dimensions. See Section

for more details.

file:///E|/BOOK/BOOK4/NODE185.HTM (4 of 5) [19/1/2003 1:31:37]

Convex Hull

Dynamic algorithms for convex-hull maintenance are data structures that permit inserting and deleting arbitrary points while always representing the current convex hull. The first such dynamic data structure [OvL81] supported insertions and deletions in time. Expositions of this result include [PS85]. Related Problems: Sorting (see page

), Voronoi diagrams (see page

).

Next: Triangulation Up: Computational Geometry Previous: Robust Geometric Primitives Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE185.HTM (5 of 5) [19/1/2003 1:31:37]

Triangulation

Next: Voronoi Diagrams Up: Computational Geometry Previous: Convex Hull

Triangulation

Input description: A set of points or a polyhedon. Problem description: Partition the interior of the point set or polyhedron into triangles. Discussion: Triangulation is a fundamental problem in computational geometry, because the first step in working with complicated geometric objects is to break them into simple geometric objects. The simplest geometric objects are triangles in two dimensions, and tetrahedra in three. Classical applications of triangulation include finite element analysis and computer graphics. A particularly interesting application of triangulation is surface or function interpolation. Suppose that we have sampled the height of a mountain at a certain number of points. How can we estimate the height at any point q in the plane? If we project the points on the plane, and then triangulate them, the triangulation completely partitions the plane into regions. We can estimate the height of q by interpolating among the three points of the triangle that contains it. Further, this triangulation and the associated height values define a surface of the mountain suitable for graphics rendering. In the plane, a triangulation is constructed by adding nonintersecting chords between the vertices until no more such chords can be added. Specific issues arising in triangulation include:

file:///E|/BOOK/BOOK4/NODE186.HTM (1 of 4) [19/1/2003 1:31:39]

Triangulation

●

●

●

●

Does the shape of the triangles in your triangulation matter? - There are usually many different ways to partition your input into triangles. Consider a set of n points in convex position in the plane. The simplest way to triangulate them would be to add to the convex hull diagonals from the first point to all of the others. However, this has the tendency to create skinny triangles. If the shape of the triangles matters for your application, you will usually want to avoid skinny triangles, or equivalently, small angles in the triangulation. The Delaunay triangulation of a point set minimizes the maximum angle over all possible triangulations. This isn't exactly what we are looking for, but it is pretty close, and the Delaunay triangulation has enough other interesting properties (including that it is dual to the Voronoi diagram) to make it the quality triangulation of choice. Further, it can be constructed in time, with implementations described below. What dimension are we working in? - As always, three-dimensional problems are harder than twodimensional problems. The three-dimensional generalization of triangulation involves partitioning the space into four-vertex tetrahedra by adding nonintersecting faces. One important difficulty is that for certain polyhedra there is no way to tetrahedralize the interior without adding extra vertices. Further, it is NP-complete to decide whether such a tetrahedralization exists, so we should not feel afraid to add extra vertices to simplify our problem. What constraints does the input have? - When we are triangulating a polygon or polyhedra, we are restricted to adding chords that do not intersect any of the boundary facets. In general, we may have a set of obstacles or constraints that cannot be intersected by inserted chords. The best such triangulation is likely to be the so-called constrained Delaunay triangulation. Implementations are described below. Are you allowed to add extra points? - When the shape of the triangles does matter, it might pay to strategically add a small number of extra ``Steiner'' points to the data set to facilitate the construction of a triangulation (say) with no small angles. As discussed above, there may be no triangulation possible for certain polyhedra without adding Steiner points.

To construct a triangulation of a convex polygon in linear time, just pick an arbitrary starting vertex v and insert chords from v to each other vertex in the polygon. Because the polygon is convex, we can be confident that none of the boundary edges of the polygon will be intersected by these chords and that all of them lie within the polygon. The simplest algorithm for constructing general polygon triangulations tries each of the possible chords and inserts them if they do not intersect a boundary edge or time and theoretically previously inserted chord. There are practical algorithms that run in interesting algorithms that run in linear time. See the implementations and notes below for details. Implementations: Triangle, by Jonathan Shewchuk of Carnegie-Mellon University, is a C language code that generates Delaunay triangulations, constrained Delaunay triangulations (forced to have certain edges), and quality-conforming Delaunay triangulations (which avoid small angles by inserting extra points). It has been widely used for finite element analysis and other applications and is fast and robust. Triangle is the first thing I would try if I needed a two-dimensional triangulation code. Although Triangle is available at http://www.cs.cmu.edu/ quake/triangle.html, it is copyrighted by the author and may not be sold or included in commercial products without a license. file:///E|/BOOK/BOOK4/NODE186.HTM (2 of 4) [19/1/2003 1:31:39]

Triangulation

GEOMPACK is a suite of Fortran 77 codes by Barry Joe of the University of Alberta, for 2- and 3dimensional triangulation and convex decomposition problems. In particular, it does both Delaunay triangulation and convex decompositions of polygonal and polyhedral regions, as well as arbitrarydimensional Delaunay triangulations. They can be obtained from ftp://ftp.cs.ualberta.ca/pub/geompack. Steve Fortune is the author of a widely used 2D code for Voronoi diagrams and Delaunay triangulations, written in C. This code is smaller and probably simpler to work with than either of the above, if all you need is the Delaunay triangulation of points in the plane. It is based on Fortune's own sweepline algorithm [For87] for Voronoi diagrams and is available from Netlib (see Section labs.com/netlib/voronoi/index.html.

) at http://netlib.bell-

) O'Rourke [O'R94] provides asymptotically slow implementations in C of polygon triangulation (in ). These will be unusable for more than modest numbers of points, and Delaunay triangulation (in but see Section

if interested. See Section

.

Algorithm 624 [Ren84] of the Collected Algorithms of the ACM is a Fortran implementation of triangulation for surface interpolation. See Section planar map is included with LEDA (see Section

. A linear-time implementation for triangulating a ).

Higher-dimensional Delaunay triangulations are a special case of higher-dimensional convex hulls, and Qhull [BDH97] appears to be the convex hull code of choice for general dimensions (i.e. three dimensions and beyond). It is written in C, and it can also construct Voronoi vertices, furthest-site Voronoi vertices, and half-space intersections. Qhull has been widely used in scientific applications and has a well-maintained home page at http://www.geom.umn.edu/software/qhull/. Notes: After a long search, Chazelle [Cha91] discovered a linear-time algorithm for triangulating a simple polygon. This algorithm is sufficiently hopeless to implement that it qualifies more as an existence proof. algorithm for polygon triangulation was given by [GJPT78]. An algorithm by The first Tarjan and Van Wyk [TW88] followed before Chazelle's result. Expositions on polygon and point set triangulation include [O'R94, PS85]. Linear-time algorithms for triangulating monotone polygons have been long known [GJPT78] and are the basis of algorithms for triangulating simple polygons. A polygon is monotone when there exists a direction d such that any line with slope d intersects the polygon in at most two points. A heavily studied class of optimal triangulations seeks to minimize the total length of the chords used. The computational complexity of constructing this minimum weight triangulation is a long-standing open problem in computational geometry, so the interest has shifted to provably good approximation time using algorithms. The minimum weight triangulation of a convex polygon can be found in file:///E|/BOOK/BOOK4/NODE186.HTM (3 of 4) [19/1/2003 1:31:39]

Triangulation

dynamic programming, as discused in Section Related Problems: Voronoi diagrams (see page

. ), polygon partitioning (see page

Next: Voronoi Diagrams Up: Computational Geometry Previous: Convex Hull Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE186.HTM (4 of 4) [19/1/2003 1:31:39]

).

Voronoi Diagrams

Next: Nearest Neighbor Search Up: Computational Geometry Previous: Triangulation

Voronoi Diagrams

Input description: A set S of points

.

Problem description: Decompose space into regions around each point such that all the points in the region around are closer to than they are to any other point in S. Discussion: Voronoi diagrams represent the region of influence around each of a given set of sites. If these sites represent the locations of McDonald's restaurants, the Voronoi diagram partitions space into cells around each restaurant. For each person living in a particular cell, the defining McDonald's represents the closest place to get a Big Mac. Voronoi diagrams have a surprising variety of uses: ●

●

Nearest neighbor search - For a query point q, finding its nearest neighbor from a fixed set of points S is simply a matter of determining which cell in the Voronoi diagram of S contains q. See for more details. Section Facility location - Suppose McDonald's wanted to open another restaurant. To minimize interference with existing McDonald's, it should be located as far away from the closest restaurant as possible. This location is always at a vertex of the Voronoi diagram, and it can be found in a

file:///E|/BOOK/BOOK4/NODE187.HTM (1 of 4) [19/1/2003 1:31:41]

Voronoi Diagrams

●

●

●

linear-time search through all the Voronoi vertices. Largest empty circle - Suppose you needed to obtain a large, contiguous, undeveloped piece of land on which to build a factory. The same condition used for picking McDonald's locations is appropriate for other undesirable facilities, namely that it be as far as possible from any relevant sites of interest. A Voronoi vertex defines the center of the largest empty circle among the points. Path planning - If the sites of S are the centers of obstacles we seek to avoid, the edges of the Voronoi diagram define the possible channels that maximize the distance to the obstacles. Thus in planning paths among the sites, it will be ``safest'' to stick to the edges of the Voronoi diagram. Quality triangulations - In triangulating a set of points, we often desire nice, fat triangles, which avoid small angles and skinny triangles. The Delaunay triangulation maximizes the minimum angle over all triangulations and is exactly what we want. Further, it is easily constructed as the dual of the Voronoi diagram. See Section

for details.

Each edge of a Voronoi diagram is part of the perpendicular bisector of two points in S, since this is the line that partitions the plane between the points. The conceptually simplest method to construct Voronoi diagrams is randomized incremental construction. To add another site to the diagram, locate the cell that contains it and add the perpendicular bisectors between the new site and all sites defining impacted regions. When the sites are inserted in random order, only a small number of regions are likely to be impacted. However, the method of choice is Fortune's sweepline algorithm, especially since robust implementations of it are readily available. Use an existing implementation instead of trying to develop your own. The algorithm works by projecting the set of sites in the plane into a set of cones in three dimensions such that the Voronoi diagram is defined by projecting the cones back onto the plane. The advantages of Fortune's algorithm are (1) it runs in optimal time, (2) it is reasonable to implement, and (3) we need not store the entire diagram if we can use it as we sweep over it. There is an interesting relationship between convex hulls in d+1 dimensions and Delaunay triangulations (or equivalently Vornoi diagrams) in d-dimensions, which provides the best way to construct Voronoi diagrams in higher dimensions. By projecting each site in into the point , taking the convex hull of this (d+1)-dimensional point set and then projecting back into d dimensions we obtain the Delaunay triangulation. Details are given in the references below. Programs that compute higher-dimensional convex hulls are discussed in Section

.

Several important variations of standard Voronoi diagrams arise in practice. See the references below for details: ●

Non-Euclidean distance metrics - Recall that the idea of a Voronoi diagram is to decompose space into regions of influence around each of the given sites. Thus far, we have assumed that Euclidean distance measures influence, but for many applications this is inappropriate. If people drive to

file:///E|/BOOK/BOOK4/NODE187.HTM (2 of 4) [19/1/2003 1:31:41]

Voronoi Diagrams

●

●

McDonald's, the time it takes to get there depends upon where the major roads are. Efficient algorithms are known for constructing Voronoi diagrams under a variety of different metrics, and for curved or constrained objects. Power diagrams - These structures decompose space into regions of influence around the sites, where the sites are no longer constrained to have all the same power. Imagine a map of the listening range of a set of radio stations operating at a given frequency. The region of influence around a station depends both on the power of its transmitter and the position of neighboring transmitters. kth-order and furthest-site diagrams - The idea of decomposing space into regions sharing some property can be taken beyond closest-point Voronoi diagrams. Any point within a single cell of the kth-order Voronoi diagram shares the same set of k closest points in S. In furthest-site diagrams, any point within a particular region shares the same furthest point in S. Point location (see Section

) on these structures permits fast retrieval of the appropriate points.

Implementations: Steve Fortune is the author of a widely used 2D code for Voronoi diagrams and Delaunay triangulations, written in C. It is based on his own sweepline algorithm [For87] for Voronoi diagrams and is likely to be the right code to try first. It is available from Netlib (see Section http://netlib.bell-labs.com/netlib/voronoi/index.html.

) at

) provides an implementation of a randomized incremental construction algorithm LEDA (see Section for planar Voronoi diagrams in C++. Higher-dimensional and furthest-site Voronoi diagrams can be constructed as a special case of higherdimensional convex hulls. Qhull [BDH97] appears to be the convex hull code of choice in three dimensions and beyond. It is written in C, and it can also construct Delaunay triangulations and halfspace intersections. Qhull has been widely used in scientific applications and has a well-maintained home page at http://www.geom.umn.edu/software/qhull/. ) contains an implementation of a randomized incremental The Stanford GraphBase (see Section algorithm to construct Voronoi diagrams and Delaunay triangulations for use as a generator of planar graph instances. Algorithm 558 [Che80] of the Collected Algorithms of the ACM is a Fortran code for the multifacility location problem. It is based on a network flow approach, instead of using Voronoi diagrams. Interestingly, the network flow code is taken from Nijenhuis and Wilf (see Section

). See Section

Notes: Voronoi diagrams were studied by Dirichlet in 1850 and are occasionally referred to as Dirichlet tessellations. They are named after G. Voronoi, who discussed them in a 1908 paper. In mathematics, concepts get named after the last person to discover them. file:///E|/BOOK/BOOK4/NODE187.HTM (3 of 4) [19/1/2003 1:31:41]

.

Voronoi Diagrams

Aurenhammer [Aur91] and Fortune [For92] provide excellent surveys on Voronoi diagrams and algorithm for constructing Voronoi associated variants such as power diagrams. The first diagrams was based on divide-and-conquer and is due to Shamos and Hoey [SH75]. Good expositions of [For87] include [O'R94]. Fortune's sweepline algorithm for constructing Voronoi diagrams in Good expositions on the relationship between Delaunay triangulations and (d+1)-dimensional convex hulls [ES86] include [O'R94]. In a kth-order Voronoi diagram, we partition the plane such that each point in a region is closest to the same set of k sites. Using the algorithm of [ES86], the complete set of kth-order Voronoi diagrams can be time. By doing point location on this structure, the k nearest neighbors to a query constructed in . Expositions on kth-order Voronoi diagrams include [O'R94, PS85]. point can be found in The smallest enclosing circle problem can be solved in time using (n-1)st order Voronoi diagrams [PS85]. In fact, there exist linear-time algorithms based on low-dimensional linear programming [Meg83]. A linear algorithm for computing the Voronoi diagram of a convex polygon is given by [AGSS89]. Related Problems: Nearest neighbor search (see page (see page

), point location (see page

).

Next: Nearest Neighbor Search Up: Computational Geometry Previous: Triangulation Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE187.HTM (4 of 4) [19/1/2003 1:31:41]

), triangulation

Nearest Neighbor Search

Next: Range Search Up: Computational Geometry Previous: Voronoi Diagrams

Nearest Neighbor Search

Input description: A set S of n points in d dimensions; a query point q. Problem description: Which point in S is closest to q? Discussion: The need to quickly find the nearest neighbor to a query point arises in a variety of geometric applications. The classic example in two dimensions is designing a system to dispatch emergency vehicles to the scene of a fire. Once the dispatcher learns the location of the fire, she uses a map to find the firehouse closest to this point so as to minimize transportation delays. This situation occurs in any application mapping customers to service providers. Nearest-neighbor search is also important in classification. Suppose we are given a collection of data about people (say age, height, weight, years of education, sex, and income level) each of whom has been labeled as Democrat or Republican. We seek a classifier to decide which way a different person is likely to vote. Each of the people in our data set is represented by a party-labeled point in d-dimensional space. A simple classifier can be built by assigning to the new point the party affiliation of its nearest neighbor.

Such nearest-neighbor classifiers are widely used, often in high-dimensional spaces. The vectorfile:///E|/BOOK/BOOK4/NODE188.HTM (1 of 4) [19/1/2003 1:31:43]

Nearest Neighbor Search

quantization method of image compression partitions an image into pixel regions. This method uses a predetermined library of several thousand pixel tiles and replaces each image region by the most similar library tile. The most similar tile is the point in 64-dimensional space that is closest to the image region in question. Compression is achieved by reporting the identifier of the closest library tile instead of the 64 pixels, at some loss of image fidelity. Issues arising in nearest-neighbor search include: ●

●

How many points are you searching? - When your data set contains only a small number of points (say ) or if only few queries are ever destined to be performed, the simple approach is best. Compare the query point q against each of the n data points. Only when fast queries are needed for large numbers of points does it pay to consider more sophisticated methods. How many dimensions are you working in? - Nearest neighbor search gets slowly but progressively harder as the dimensionality increases. The kd-tree data structure, presented in , does a very good job in moderate-dimensional spaces, even the plane. Still, above 20 Section or so dimensions, you might as well do a linear search through the data points. Search in highdimensional spaces becomes hard because a sphere of radius r, representing all the points with from the center, progressively fills up less volume relative to a cube as the distance dimensionality increases. Thus any data structure based on partitioning points into enclosing volumes will become less and less effective.

●

) provide an efficient data structure for In two dimensions, Voronoi diagrams (see Section nearest-neighbor queries. The Voronoi diagram of a point set in the plane decomposes the plane into regions such that the cell containing each data point consists of the part of the plane that is nearer to that point than any other in the set. Finding the nearest neighbor of query point q reduces to finding which cell in the Voronoi diagram contains q and reporting the data point associated with it. Although Voronoi diagrams can be built in higher dimensions, their size rapidly grows to the point of unusability. Is your data set static or dynamic? - Will there be occasional insertions or deletions of new data points in your application? If these are just rare events, it might pay to build your data structure from scratch each time. If they are frequent, select a version of the kd-tree that supports insertions and deletions.

The nearest neighbor graph on a set S of n points links each vertex to its nearest neighbor. This graph is a subgraph of the Delaunay triangulation and so can be computed in . This is quite a bargain since it takes time just to discover the closest pair of points in S. As a lower bound, the closest pair problem in one dimension reduces to sorting. In a sorted set of numbers, the closest pair corresponds to two numbers that lie next to each other in sorted order, so we need only check which is the minimum gap between the n-1 adjacent pairs. The limiting case of this occurs when the closest pair are distance zero apart, meaning that the elements are not unique. file:///E|/BOOK/BOOK4/NODE188.HTM (2 of 4) [19/1/2003 1:31:43]

Nearest Neighbor Search

Implementations: Ranger is a tool for visualizing and experimenting with nearest-neighbor and orthogonal-range queries in high-dimensional data sets, using multidimensional search trees. Four different search data structures are supported by Ranger: naive kd-trees, median kd-trees, nonorthogonal kd-trees, and the vantage point tree. For each of these, Ranger supports queries in up to 25 dimensions under any Minkowski metric. It includes generators for a variety of point distributions in arbitrary dimensions. Finally, Ranger provides a number of features to aid in visualizing multidimensional data, best illustrated by the accompanying video [MS93]. To identify the most appropriate projection at a glance, Ranger provides a matrix of all two-dimensional projections of the data set. Ranger is written in C, using Motif. It runs on Silicon Graphics and HP workstations and is available in the algorithm repository http://www.cs.sunysb.edu/ algorith. See Section

for a complete collection of Voronoi diagram implementations. In particular, LEDA (see

) provides an implementation of 2D Voronoi diagrams in C++, as well as planar point Section location to make effective use of them for nearest-neighbor search. A Pascal implementation of the divide-and-conquer algorithm for finding the closest pair of points in a set of n points appears in [MS91]. See Section

.

Notes: The best reference on kd-trees and other spatial data structures is two volumes by Samet [Sam90b, Sam90a], where all major variants are developed in substantial detail. Good expositions on finding the closest pair of points in the plane [BS76] include [CLR90, Man89]. These algorithms use a divide-and-conquer approach instead of just selecting from the Delaunay triangulation. A recent development in higher-dimensional nearest-neighbor search is algorithms that quickly produce a point that if not the nearest neighbor lies provably close to the query point. A sparse weighted graph structure is built from the data set, and the nearest neighbor is found by starting at a random point and greedily walking in the graph towards the query point. The closest point found over several random trials is declared the winner. Similar data structures hold promise for other problems in high-dimensional spaces. See [AM93, AMN 94]. Related Problems: Kd-trees (see page

), Voronoi diagrams (see page

).

file:///E|/BOOK/BOOK4/NODE188.HTM (3 of 4) [19/1/2003 1:31:43]

), range search (see page

Nearest Neighbor Search

Next: Range Search Up: Computational Geometry Previous: Voronoi Diagrams Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE188.HTM (4 of 4) [19/1/2003 1:31:43]

Range Search

Next: Point Location Up: Computational Geometry Previous: Nearest Neighbor Search

Range Search

Input description: A set S of n points in

and a query region Q.

Problem description: Which points from S lie within Q? Discussion: Range search problems arise in database and geographic information system (GIS) applications. Any data object with d numerical fields, such as person with height, weight, and income, can be modeled as a point in d-dimensional space. A range query describes a region in space and asks for all points or the number of points in the region. For example, asking for all people with income between $0 and $10,000, with height between 6'0'' and 7'0'', and weight between 50 and 140 lbs. defines a box containing people whose body and wallets are both thin. The difficulty of a range search problem depends on several factors: ●

●

How many range queries are you going to perform? - The simplest approach to range search tests each of the n points one by one against the query polygon Q. This works just fine if the number of queries will be small. Algorithms to test whether a point is within a given polygon are presented . in Section What shape is your query polygon? - The easiest type of regions to query against are axis-parallel

file:///E|/BOOK/BOOK4/NODE189.HTM (1 of 3) [19/1/2003 1:31:45]

Range Search

rectangles, because the inside/outside test reduces to verifying whether each coordinate lies within a prescribed range. The figure above illustrates such an orthogonal range query. If you are querying against a nonconvex polygon, it will pay to partition your polygon into convex pieces or (even better) triangles and then query the point set against each one of the pieces. This is because testing whether a point is inside a convex polygon can be done more quickly and easily than for arbitrary polygons. Algorithms for such convex decompositions are discussed in Section ●

●

. How many dimensions? - The best general-purpose approach to range queries builds a kd-tree on . To perform a query, a depth-first traversal of the kd-tree the point set, as discussed in Section is performed, with each tree node expanded only if the associated rectangle intersects the query region. For sufficiently large or misaligned query regions, the entire tree might have to be traversed, but in general; kd-trees lead to an efficient solution. Although algorithms with efficient worst-case performance are known in two dimensions, kd-trees are likely to work even better in the plane. In higher-dimensions, kd-trees provide the only viable solution to the problem. Can I just count the number of points in a region, or do I have to identify them? - For many applications it suffices to count the number of points in a region instead of returning them. Harkening back to the introductory example, we may want to know whether there are more thin/poor people or rich/fat ones. The need to find the densest or emptiest region in space often naturally arises, and the problem can be solved by counting range queries. A data structure for efficiently answering such aggregate range queries can be based on the dominance ordering of the point set. A point x is said to dominate point y if y lies both below and to the left of x. Let DOM(p) be a function that counts the number of points in S that are dominated by p. The number of points m in the orthogonal rectangle defined by and is given by

The second additive term corrects for the points for the lower left-hand corner that have been subtracted away twice. To answer arbitrary dominance queries efficiently, partition the space into rectangles by drawing a horizontal and vertical line through each of the n points. The set of points dominated is identical for each point in each rectangle, so the dominance count of the lower left-hand corner of each rectangle can precomputed, stored, and reported for any query point within it. Queries reduce to binary search and thus take time. Unfortunately, this data structure takes quadratic space. However, the same idea can be adapted to kd-trees to create a more space-efficient search structure. Implementations: LEDA (see Section

) provides excellent support for maintaining planar

file:///E|/BOOK/BOOK4/NODE189.HTM (2 of 3) [19/1/2003 1:31:45]

Range Search

subdivisions in C++. In particular, it supports orthogonal range queries in time, where n is the complexity of the subdivision and k is the number of points in the rectangular region. Ranger is a tool for visualizing and experimenting with nearest-neighbor and orthogonal-range queries in high-dimensional data sets, using multidimensional search trees. Four different search data structures are supported by Ranger: naive kd-trees, median kd-trees, nonorthogonal kd-trees, and the vantage point tree. For each of these, Ranger supports queries in up to 25 dimensions under any Minkowski metric. It is available in the algorithm repository. A bare bones implementation in C of orthogonal range search using kd-trees appears in [GBY91]. Sedgewick [Sed92] provides code fragments of the grid method for orthogonal range search in C++. See Section

for details on both of them.

Notes: Good expositions on data structures with worst-case performance for orthogonal-range searching [Wil85] include [PS85]. An exposition on kd-trees for orthogonal range queries in two dimensions appears in [PS85]. Their worst-case performance can be very bad; [LW77] describes an time to report that a rectangle is empty. instance in two dimensions requiring The problem becomes considerably more difficult for nonorthogonal range queries, where the query region is not an axis-aligned rectangle. For half-plane intersection queries, time and linear space suffice [CGL85]; for range searching with simplex query regions (such as a triangle in the plane), lower bounds preclude efficient worst-case data structures. See [Yao90] for a discussion. Related Problems: Kd-trees (see page

), point location (see page

),

Next: Point Location Up: Computational Geometry Previous: Nearest Neighbor Search Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE189.HTM (3 of 3) [19/1/2003 1:31:45]

Point Location

Next: Intersection Detection Up: Computational Geometry Previous: Range Search

Point Location

Input description: A decomposition of the plane into polygonal regions and a query point q. Problem description: Which region contains the query point q? Discussion: Point location is a fundamental subproblem in computational geometry, usually needed as an ingredient to solve larger geometric problems. In a dispatch system to assign policemen to the scene of a crime, the city will be partitioned into different precincts or districts. Given a map of regions and a query point (the crime scene), the system must identify which region contains the point. This is exactly the problem of planar point location, variations of which include: ●

Is a given point inside or outside of polygon P? - The simplest version of point location involves only two regions, inside-P and outside-P, and asks which contains a given query point. For polygons with lots of narrow spirals, this can be surprisingly difficult to tell by inspection. The secret to doing it both by eye or machine is to draw a ray starting from the query point and ending beyond the furthest extent of the polygon. Count the number of times the polygon crosses through an edge. If this number is odd, we must be within the polygon. If it is even, we must be outside. The case of the line passing through a vertex instead of an edge is evident from context, since we are counting the number of times we pass through the boundary of the polygon. Testing each of

file:///E|/BOOK/BOOK4/NODE190.HTM (1 of 4) [19/1/2003 1:31:46]

Point Location

●

●

●

●

the n edges for intersection against the query ray takes O(n) time. Faster algorithms for convex polygons are based on binary search and take time. How many queries will have to be performed? - When we have a subdivision with multiple regions, it is always possible to repeat the inside-polygon test above on each region in the subdivision. However, this is wasteful if we will be performing many such point location queries on the same subdivision. Instead, we can construct a grid-like or tree-like data structure on top of our subdivision to get us near the correct region quickly. Such search structures are discussed in more detail below. How complicated are the regions of your subdivision? - More sophisticated inside-outside tests are required when the regions of your subdivision are arbitrary polygons. By triangulating all polygonal regions first, each inside-outside test reduces to testing whether a point is in a triangle. Such a test can be made particularly fast and simple, at the minor cost of recording the fullpolygon name for each triangle. An added benefit is that the smaller your regions are, the better grid-like or tree-like superstructures are likely to perform. Some care should be taken when you . triangulate to avoid long skinny triangles, as discussed in Section How regularly sized and spaced are your regions? - If all resulting triangles are about the same size and shape, the simplest point location method imposes a regularly-spaced grid of horizontal and vertical lines over the entire subdivision. For each of the rectangular regions, we maintain a list of all the regions that are at least partially contained within the rectangle. Performing a point location query in such a grid file involves a binary search or hash table lookup to identify which rectangle contains query point q and then searching each region in the resulting list to identify the right one. Such grid files will perform very well, provided that each triangular region overlaps only a few rectangles (thus minimizing storage space) and each rectangle overlaps only a few triangles (thus minimizing search time). Whether it will perform well is a function of the regularity of your subdivision. Some flexibility can be achieved by spacing the horizontal lines irregularly, as a function of the regions of the subdivision. The slab method, discussed below, is a variation on this idea that guarantees worst-case efficient point location at the cost of quadratic space. How many dimensions will you be working in? - In three or more dimensions, some flavor of kdtree will almost certainly be the point-location method of choice. They are also likely to be the right answer for planar subdivisions too irregular for grid files. Kd-trees, described in Section , decompose the space into a hierarchy of rectangular boxes. At each node in the tree, the current box is split into a small number (typically 2 or 4 or , where d is the dimension) of smaller boxes. At the leaves of the tree, each box is labeled with the small number of regions that are at least partially contained in the box. The point location search starts at the root of the tree and keeps traversing down the child whose box contains the query point. When the search hits a leaf, we test each of the relevant regions against q to see which one of them contains the point. As with grid files, we hope that each leaf contains a small number of regions and that each region does not cut across too many leaf cells.

file:///E|/BOOK/BOOK4/NODE190.HTM (2 of 4) [19/1/2003 1:31:46]

Point Location

The simplest algorithm to guarantee worst-case access is the slab method, which draws horizontal lines through each vertex, thus creating n+1 ``slabs'' between the lines. Since the slabs are defined by horizontal lines, finding the slab containing a particular query point can be done using a binary search on the y-coordinate of q. Since there can be no vertices within any slab, looking for the region containing a point within a slab can be done by a second binary search on the edges that cross the slab. The catch is that a binary search tree must be maintained for each slab, for a worst-case of space if each region intersects each slab. A more space-efficient approach based on building a hierarchy of triangulations over for search and is discussed in the notes below. the regions also achieves Worst-case efficient computational geometry methods either require a lot of storage or are fairly complicated to implement. We identify implementations of worst-case methods below, which are worth at least experimenting with. However, we recommend kd-trees for most general point-location applications. Implementations: LEDA (see Section ) provides excellent support for maintaining planar time. subdivisions in C++ and, in particular, supports point location in Arrange is a package for maintaining arrangements of polygons in either the plane or on the sphere. Polygons may be degenerate, and hence represent arrangements of lines. A randomized incremental construction algorithm is used, and efficient point location on the arrangement is supported. Polygons may be inserted but not deleted from the arrangement, and arrangements of several thousand vertices and edges can be constructed in a few seconds. Arrange is written in C by Michael Goldwasser and is available from http://theory.stanford.edu/people/wass/wass.html. A routine in C to test whether a point lies in a simple polygon has been provided by O'Rourke [O'R94], and a Pascal routine for the same problem by [MS91]. For information on both, see Section

.

Notes: The inside-outside test for convex polygons is described in [PS85], which has a very thorough treatment of deterministic planar point location data structures. Expositions on the inside-outside test for simple polygons include [Man89, PS85]. An experimental study of algorithms for planar point location is described in [EKA84]. The winner was a bucketing technique akin to the grid file. The elegant triangle refinement method of Kirkpatrick [Kir83] builds a hierarchy of triangulations above the actual planar subdivision such that each triangle on a given level intersects only a constant number of triangles on the following level. Since each triangulation is a fraction of the size of the subsequent one, the total space is obtained by summing up a geometric series and hence is linear. Further, the height of , ensuring fast query times. An alternative algorithm realizing the same time the hierarchy is bounds is [EGS86]. The slab method described above is due to [DL76] and is presented in [PS85]. file:///E|/BOOK/BOOK4/NODE190.HTM (3 of 4) [19/1/2003 1:31:46]

Point Location

More recently, there has been interest in dynamic data structures for point location, which support fast incremental updates of the planar subdivision (such as insertions and deletions of edges and vertices) as well as fast point location. Chiang and Tamassia's [CT92] survey is an appropriate place to begin. Related Problems: Kd-trees (see page (see page

), Voronoi diagrams (see page

), nearest neighbor search

).

Next: Intersection Detection Up: Computational Geometry Previous: Range Search Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE190.HTM (4 of 4) [19/1/2003 1:31:46]

Intersection Detection

Next: Bin Packing Up: Computational Geometry Previous: Point Location

Intersection Detection

Input description: A set S of lines and line segments .

or a pair of polygons or polyhedra

and

Problem description: Which pairs of line segments intersect each other? What is the intersection of and ? Discussion: Intersection detection is a fundamental geometric primitive that arises in many applications. Picture a virtual-reality simulation of an architectural model for a building. The illusion of reality vanishes the instant the virtual person walks through a virtual wall. To enforce such physical constraints, any such intersection between polyhedral models must be immediately detected and the operator notified or constrained. Another application of intersection detection is design rule checking for VLSI layouts. A minor design mistake resulting in two crossing metal strips can short out the chip, but such errors can be detected before fabrication using programs to find all intersections between line segments. Issues arising in intersection detection include: file:///E|/BOOK/BOOK4/NODE191.HTM (1 of 5) [19/1/2003 1:31:48]

Intersection Detection

●

●

Do you want to compute the intersection or just report it? - We distinguish between intersection detection and computing the actual intersection. The latter problem is obviously harder than the former and is not always necessary. In the virtual-reality application above, it might not be important to know exactly where we hit the wall, just that we hit it. Are you intersecting lines or line segments? - The big difference here is that any two lines with different slopes must intersect at exactly one point. By comparing each line against every other line, all points of intersections can be found in constant time per point, which is clearly optimal. Constructing the arrangement of the lines provides more information than just intersection points and is discussed in Section

.

Finding all the intersections between n line segments is considerably more challenging. Even the basic primitive of testing whether two line segments intersect is not as trivial as it might seem,

●

●

●

. To find all intersections, we can explicitly test each line and this is discussed in Section segment against each other line segment and thus find all intersections in time. Each segment can always intersect each other segment, yielding a quadratic number of intersections, so in the worst case, this is optimal. For many applications, however, this worst case is not very interesting. How many intersection points do you expect? - Sometimes, as in VLSI design rule checking, we expect the set of line segments to have few if any intersections. What we seek is an algorithm whose time is output sensitive, taking time proportional to the number of intersection points. Such output-sensitive algorithms exist for line-segment intersection, with the fastest algorithm taking time, where k is the number of intersections. Such algorithms are sufficiently complicated that you should use an existing implementation if you can. These algorithms are based on the sweepline approach, discussed below. Can you see point x from point y? - Visibility queries ask whether vertex x can see vertex y unobstructed in a room full of obstacles. This can be phrased as the following line-segment intersection problem: does the line segment from x to y intersect any obstacle? Such visibility ) and in hidden-surface elimination for problems arise in robot motion planning (see Section computer graphics. Are the intersecting objects convex? - Better intersection algorithms exist when the line segments form the boundaries of polygons. The critical issue here becomes whether the polygons are convex. Intersecting a convex n-gon with a convex m-gon can be done in O(n+m) time, using a sweepline algorithm as discussed below. This is possible because the intersection of two convex polygons must form another convex polygon using at most n+m vertices. However, the intersection of two nonconvex polygons is not so well behaved. Consider the intersection of two ``combs'' generalizing the Picasso-like frontspiece to this section. As illustrated, the intersection of nonconvex polygons may be disconnected and have quadratic size in the worst case.

file:///E|/BOOK/BOOK4/NODE191.HTM (2 of 5) [19/1/2003 1:31:48]

Intersection Detection

●

Intersecting two polyhedra is somewhat more complicated than intersecting polygons, because two polyhedra can intersect even when no edges do. Consider the example of a needle piercing the interior of a face. In general, however, the same issues arise for both polygons and polyhedra. Are you searching for intersections repeatedly with the same basic objects? - In the walk-through application described above, the room and the objects in it don't change between one scene and the next. Only the person moves, and, further, the intersections are rare. One common solution is to approximate the objects in the scene by simpler objects that enclose them, such as boxes. Whenever two enclosing boxes intersect, then the underlying objects might intersect, and so further work is necessary to decide the issue. However, it is much more efficient to test whether simple boxes intersect than more complicated objects, so we win if collisions are rare. Many variations on this theme are possible, but this idea can lead to large performance improvements for complicated environments.

Planar sweep algorithms can be used to efficiently compute the intersections among a set of line segments, or the intersection/union of two polygons. These algorithms keep track of interesting changes as we sweep a vertical line from left to right over the data. At its leftmost position, the line intersects nothing, but as it moves to the right, it encounters a series of events: ●

●

●

Insertion - the leftmost point of a line segment may be encountered, and it is now available to intersect some other line segment. Deletion - the rightmost point of a line segment is encountered. This means that we have completely swept over the segment on our journey, and so it can be deleted from further consideration. Intersection - if we maintain the active line segments that intersect the sweep line as sorted from top to bottom, the next intersection must occur between neighboring line segments. After the intersection, these two line segments swap their relative order.

Keeping track of what is going on requires two data structures. The future is maintained by an event queue, a priority queue ordered by the x-coordinate of all possible future events of interest: insertion, deletion, and intersection. See Section for priority queue implementations. The present is represented by the horizon, an ordered list of line segments intersecting the current position of the sweepline. The horizon can be maintained using any dictionary data structure, such as a balanced tree. To adapt this approach to computing the intersection or union of polygons, we modify the processing of the three event types to keep track of what has occurred to the left of the sweepline. This algorithm can be considerably simplified for pairs of convex polygons, since (1) at most four polygon edges intersect the sweepline, so no horizon data structure is needed and (2) no event-queue sorting is needed, since we can start from the leftmost vertex of each polygon and proceed to the right by following the polygonal ordering.

file:///E|/BOOK/BOOK4/NODE191.HTM (3 of 5) [19/1/2003 1:31:48]

Intersection Detection

Implementations: LEDA (see Section ) provides a C++ implementation of the Bentley-Ottmann sweepline algorithm [BO79], finding all k intersection points between n line segments in the plane in time. O'Rourke [O'R94] provides a robust program in C to compute the intersection of two convex polygons. See Section

.

RAPID is a ``Rapid and Accurate Polygon Interference Detection library'' for large environments composed of polygonal models. It is free for noncommercial use and available from http://www.cs.unc.edu/ geom/OBB/OBBT.html. Model Pascal subroutines for convexity testing and for finding an intersection in a set of line segments appear in [MS91]. See Section for details. XTango (see Section polygon clipping against a polygonal window.

) includes an animation of

Finding the mutual intersection of a collection of half-spaces is a special case of higher-dimensional convex hulls, and Qhull [BDH97] is convex hull code of choice for general dimensions. Qhull has been widely used in scientific applications and has a well-maintained home page at http://www.geom.umn.edu/software/qhull/. Notes: Good expositions on line segment intersection detection [BO79, SH75] include [CLR90, Man89, NH93, PS85]. Good expositions on polygon and polyhedra intersection [HMMN84, MP78, SH76] include [PS85]. Preparata and Shamos [PS85] provide a good exposition on the special case of finding intersections and unions of axis-oriented rectangles, a problem that arises often in VLSI design. An optimal algorithm for computing line segment intersections is due to Chazelle and Edelsbrunner [CE92]. Simpler, randomized algorithms achieving the same time bound are thoroughly presented by Mulmuley [Mul94]. Surveys on hidden-surface removal include [Dor94, SSS74]. Related Problems: Maintaining arrangements (see page

), motion planning (see page

Next: Bin Packing Up: Computational Geometry Previous: Point Location file:///E|/BOOK/BOOK4/NODE191.HTM (4 of 5) [19/1/2003 1:31:48]

).

Intersection Detection

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK4/NODE191.HTM (5 of 5) [19/1/2003 1:31:48]

Bin Packing

Next: Medial-Axis Transformation Up: Computational Geometry Previous: Intersection Detection

Bin Packing

Input description: A set of n items with sizes

. A set of m bins with capacity

.

Problem description: How can you store all the items using the smallest number of bins? Discussion: Bin packing arises in a variety of packaging and manufacturing problems. Suppose that you are manufacturing widgets with parts cut from sheet metal, or pants with parts cut from cloth. To minimize cost and waste, we seek to lay out the parts so as to use as few fixed-size metal sheets or bolts of cloth as possible. Identifying which part goes on which sheet in which location is a bin-packing variant called the cutting stock problem. After our widgets have been successfully manufactured, we will be faced with another bin packing problem, namely how best to fit the boxes into trucks to minimize the number of trucks needed to ship everything. Even the most elementary-sounding bin-packing problems are NP-complete (see the discussion of integer partition in Section ), so we are doomed to think in terms of heuristics instead of worst-case optimal algorithms. Fortunately, relatively simple heuristics tend to work well on most bin-packing problems. Further, many applications have peculiar, problem-specific constraints that would frustrate highly tuned algorithms for the problem. The following factors will affect the choice of heuristic:

file:///E|/BOOK/BOOK5/NODE192.HTM (1 of 4) [19/1/2003 1:31:50]

Bin Packing ●

●

●

What are the shapes and sizes of objects? - The character of the problem depends greatly on the shapes of the objects to be packed. Packing pieces of a standard jigsaw puzzle is a different problem than packing squares into a rectangular box. In one-dimensional bin packing, each object's size is given simply as an integer, making it a version of the knapsack problem of Section . This is equivalent to packing boxes of equal width into a chimney of that width. If all the boxes are of identical size and shape, the optimal packing is simply a cubic lattice, oriented appropriately with the walls of the bin. It is a waste to use the methods described in this section on such simple problems. Are there constraints on the orientation and placement of objects? - In loading a truck, many of the boxes will say ``this side up.'' Respecting this constraint restricts our flexibility in packing and will likely lead to an increase in the number of trucks needed to send out the shipment. Similarly, boxes with fragile objects may be labeled ``do not stack'' and thus are constrained to sit on top of a pile of boxes. Most shippers seem to ignore these constraints. Indeed, your task will be simpler if you don't have to worry about the consequences of them. Is the problem on-line or off-line? - Do we know the complete set of objects that we will be packing at the beginning of the job (an off-line problem)? Or will we get them one at a time and have to deal with them as they arrive (an on-line problem)? The difference is important, because we can do a better job packing when we can take a global view and plan ahead. For example, we can arrange the objects in an order that will facilitate efficient packing, perhaps by sorting them from biggest to smallest.

The standard heuristics for bin packing order the objects by size or shape (or in an on-line problem, simply the order they arrive in) and then insert them into bins. Typical insertion rules include (1) select the first or leftmost bin the object fits in, (2) select the bin with the most room, (3) select the bin that provides the tightest fit, or (4) select a random bin. Analytical and empirical results suggest that the best heuristic is first-fit, decreasing. Sort the objects in decreasing order of size, so that the biggest object is first and the smallest last. Insert each object one by one into the first bin that has room for it. If no bin has room for it, we must start another bin. In the case of one-dimensional bin packing, this can never require more than 22% more bins than necessary and usually does much better. First-fit decreasing has an intuitive appeal to it; we pack the bulky objects first and hope that the little objects can be used to fill up the cracks. First-fit decreasing is easily implemented in time, where is the number of bins actually used, by doing a linear sweep through the bins on each insertion. A faster implementation is possible by using binary tree to keep track of the space remaining in each bin. We can fiddle with the insertion order in such a scheme to deal with problem-specific constraints. For example, it is reasonable to take ``do not stack'' boxes last (perhaps after artificially lowering the height of the bins to give some room up top to work with) and to place fixed-orientation boxes at the beginning (so we can use the extra flexibility later to stick boxes into cracks). Packing boxes is much easier than packing arbitrary geometric shapes, enough so that a reasonable file:///E|/BOOK/BOOK5/NODE192.HTM (2 of 4) [19/1/2003 1:31:50]

Bin Packing

approach for general shapes is to pack each part into its own box and then pack the boxes. Finding an enclosing rectangle for a polygonal part is easy; just find the upper, lower, left, and right tangents in a given orientation. A minimum-area enclosing rectangle can be found by determining the orientation that leads to the smallest box. In the case of nonconvex parts, considerable useful space can be wasted in the holes created by placing the part in a box. One solution is to find the maximum empty rectangle within each boxed part and use this to contain other parts if it is sufficiently large. More advanced solutions are discussed in the references. Implementations: Codes for the one-dimensional version of bin packing, the so-called knapsack problem, are presented in Section

.

XTango (see Section ) includes an animation of the first-fit bin packing heuristic. Test data for bin packing is available from http://mscmga.ms.ic.ac.uk/info.html. Notes: See [CGJ96] for a survey of the extensive literature on approximation algorithms for bin packing. Expositions on heuristics for bin packing include [Baa88]. Experimental results on bin-packing heuristics include [BJLM83]. Sphere packing is an important and well-studied special case of bin packing, with applications to errorcorrecting codes. Particularly notorious is the ``Kepler conjecture,'' the apparently still-open problem of establishing the densest packing of unit spheres in three dimensions. Conway and Sloane [CS93] is the best reference on sphere packing and related problems. Sloane provides an extensive set of tables of the best known packings, available from ftp://netlib.bell-labs.com. Victor Milenkovic and his students have worked extensively on two-dimensional bin-packing problems for the apparel industry, minimizing the amount of material needed to manufacture pants and other clothing. Recent reports of this work include [DM97, Mil97]. Related Problems: Knapsack problem (see page

), set packing (see page

).

Next: Medial-Axis Transformation Up: Computational Geometry Previous: Intersection Detection Algorithms file:///E|/BOOK/BOOK5/NODE192.HTM (3 of 4) [19/1/2003 1:31:50]

Bin Packing

Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE192.HTM (4 of 4) [19/1/2003 1:31:50]

Medial-Axis Transformation

Next: Polygon Partitioning Up: Computational Geometry Previous: Bin Packing

Medial-Axis Transformation

Input description: A polygon or polyhedron P. Problem description: What is the set of points within P that have more than one closest point on the boundary of P? Discussion: The medial-axis transformation is useful in thinning a polygon, or, as is sometimes said, finding its skeleton. The goal is to extract a simple, robust representation of the shape of the polygon. As can be seen from the figures above, the thinned versions of the letters capture the essence of the shape of an `A' and a `B', and would be relatively unaffected by changing the thickness of strokes or by adding font-dependent flourishes such as serifs. The medial-axis transformation of a polygon is always a tree, making it fairly easy to use dynamic programming to measure the ``edit distance'' between the skeleton of a known model and the skeleton of an unknown object. Whenever the two skeletons are close enough, we can classify the unknown object as an instance of our model. This technique has proven useful in computer vision and in optical character recognition. The skeleton of a polygon with holes (like the `A' and `B') is not a tree, but an embedded planar graph, but it remains easy to work with.

file:///E|/BOOK/BOOK5/NODE193.HTM (1 of 3) [19/1/2003 1:31:52]

Medial-Axis Transformation

There are two distinct approaches to computing medial-axis transforms, depending upon whether your inputs are arbitrary geometric points or pixel images: ●

Geometric data - Recall that the Voronoi diagram of a point set S decomposes the plane into such that each point within the region around is closer to regions around each point than to any other site in S. Similarly, the Voronoi diagram of a set of line segments L decomposes such that each point within the region the plane into regions around each line segment around is closer to than to any other site in L. . The Any polygon is defined by a collection of line segments such that shares a vertex with medial-axis transform of a polygon P is simply the portion of the line-segment Voronoi diagram that lies within P.

●

Any line-segment Voronoi diagram code thus suffices to do polygon thinning. In the absence of such a code, the most readily implementable thinning algorithm starts at each vertex of the polygon and grows the skeleton inward with an edge bisecting the angle between the two neighboring edges. Eventually, these two edges meet at an internal vertex, a point equally far from three line segments. One of the three is now enclosed within a cell, while a bisector of the two surviving segments grows out from the internal vertex. This process repeats until all edges terminate in vertices. Image data - Whenever attempting geometric approaches to image processing problems, we must remain aware that images are pixel-based and not continuous. All the pixels sit as lattice points on an integer grid. While we can extract a polygonal description from the boundary and feed it to the geometric algorithms above, the internal vertices of the skeleton will most likely not lie at grid points. This may well make geometric approaches inappropriate for your intended application. Algorithms that explicitly manipulate pixels tend to be easy to implement, because they avoid complicated data structures. The basic pixel-based approach for constructing a skeleton directly implements the ``brush fire'' view of thinning. Imagine a brush fire along all edges of the polygon, burning inward at a constant speed. The skeleton is marked by all points where two or more fires meet. The resulting algorithm traverses all the boundary pixels of the object, deletes all except the extremal pixels, and repeats. The algorithm terminates when all pixels are extreme, leaving an object only 1 or 2 pixels thick. When implemented properly, this takes time linear in the number of pixels in the image. The trouble with such pixel-based approaches is that the geometry doesn't work out exactly right. For example, the skeleton of a polygon is no longer always a tree or even necessarily connected, and the points in the skeleton will be close-to-but-not-quite equidistant to two boundary edges. The usual solution is to tweak your implementation until it gives skeletons that look decent on the data that you think is most interesting. Since you are trying to do continuous geometry in a discrete world, there is no way to solve the problem completely. You just have to live with it.

file:///E|/BOOK/BOOK5/NODE193.HTM (2 of 3) [19/1/2003 1:31:52]

Medial-Axis Transformation

Implementations: MAT [Ogn93] is a medial-axis transform code designed for 2D skeletonization of binary images, written by Robert Ogniewicz. MAT accepts a variety of different input formats, including polygonal representations. This seems to be a solidly built program, and it should be your first stop on seeking a routine for thinning. It available from http://hrl.harvard.edu/people/postdocs/rlo/rlo.dir/rlosoft.html. Programs for constructing Voronoi diagrams are discussed in Section

.

Notes: For a comprehensive surveys of thinning approaches in image processing, see [LLS92, Ogn93]. The medial axis transformation was introduced for shape similarity studies in biology [Blu67]. Applications of the medial-axis transformation in pattern recognition are discussed in [DH73]. Good expositions on the medial-axis transform include [O'R94, Pav82]. time for arbitrary n-gons [Lee82], although The medial-axis of a polygon can be computed in algorithm for constructing linear-time algorithms exist for convex polygons [AGSS89]. An medial-axis transforms in curved regions was given by Kirkpatrick [Kir79]. Related Problems: Voronoi diagrams (see page

), Minkowski sum (see page

Next: Polygon Partitioning Up: Computational Geometry Previous: Bin Packing Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE193.HTM (3 of 3) [19/1/2003 1:31:52]

).

Polygon Partitioning

Next: Simplifying Polygons Up: Computational Geometry Previous: Medial-Axis Transformation

Polygon Partitioning

Input description: A polygon or polyhedron P. Problem description: How can P be partitioned into a small number of simple (typically convex) pieces? Discussion: Polygon partitioning is an important preprocessing step for many geometric algorithms, because most geometric problems are simpler and faster on convex objects than on nonconvex ones. We are better off whenever we can partition a nonconvex object into a small number of convex pieces, because it is easier to work with the pieces independently than with the original object. Several flavors of polygon partitioning arise, depending upon the particular application: ●

Should all the pieces be triangles? - The mother of all polygon partitioning problems is triangulation, where the interior of the polygon is completely partitioned into triangles. Triangles are always convex and have only three sides, so any geometric operation performed on a triangle is destined to be as simple as it can be. Because triangles all have three sides, all triangulations of polygons contain exactly the same

file:///E|/BOOK/BOOK5/NODE194.HTM (1 of 3) [19/1/2003 1:31:53]

Polygon Partitioning

number of pieces. Therefore, triangulation cannot be the answer if we seek a small number of convex pieces. The goal of finding ``nice'' triangulations revolves around the shape of the ●

●

for a thorough discussion of triangulation. triangles. See Section Do I want to cover or partition my polygon? - Partitioning a polygon means completely dividing the interior into nonoverlapping pieces. Covering a polygon means that our decomposition is permitted to contain mutually overlapping pieces. Both can be useful in different situations. In ), we seek decomposing a complicated query polygon in preparation for range search (Section a partitioning, so that each point we locate occurs in exactly one piece. In decomposing a polygon for painting purposes, a covering suffices, since there is no difficulty with filling in a region twice. We will concentrate here on partitioning, since it is simpler to do right, and any application needing a covering will accept a partitioning. The only potential drawback is that partitions can be somewhat larger than coverings. Am I allowed to add extra vertices? - A final issue associated with polygon decomposition is whether we are allowed to add Steiner vertices (either by splitting edges or adding interior points) or whether we are restricted to adding chords between two existing vertices. The former may result in a smaller number of pieces, at the cost of more complicated algorithms and perhaps messier results.

The Hertel-Mehlhorn heuristic for convex decomposition using diagonals is simple, efficient, and always produces no more than four times the minimum number of convex pieces. It starts with an arbitrary triangulation of the polygon and then deletes a chord that leaves only convex pieces. The decision of whether a chord deletion will create a nonconvex piece can be made locally from the chords and edges surrounding the chord, in constant time. A vertex in a polygon is reflex if the internal angle is greater than 180 degrees. We can delete any chord that does not create a reflex vertex. I recommend using this heuristic unless it is critical for you to absolutely minimize the number of pieces. By experimenting with different triangulations and various deletion orders, you may be able to obtain somewhat better decompositions. Dynamic programming may be used to minimize the number of diagonals used in the decomposition. The simplest implementation, which maintains the number of pieces for all subpolygons split by a . Faster algorithms use fancier data structures, running in time, where r is chord, runs in algorithm that further reduces the number of pieces by adding the number of reflex vertices. An interior vertices is cited below, although it is complex and presumably difficult to implement. Implementations: Many triangulation codes start by finding a trapezoidal or monotone decomposition of polygons. Further, a triangulation is a simple form of convex decomposition. Check out the codes in Section

as a starting point for most any decomposition problem.

A triangulation code of particular relevance here is GEOMPACK, a suite of Fortran 77 codes by Barry Joe, for 2- and 3-dimensional triangulation and convex decomposition problems. In particular, it does file:///E|/BOOK/BOOK5/NODE194.HTM (2 of 3) [19/1/2003 1:31:53]

Polygon Partitioning

both Delaunay triangulation and convex decompositions of polygonal and polyhedral regions, as well as arbitrary-dimensional Delaunay triangulations. Notes: Keil and Sack [KS85] given an excellent survey on what is known about partitioning and covering polygons. Expositions on the Hertel-Mehlhorn heuristic [HM83] include [O'R94]. The dynamic programming algorithm for minimum convex decomposition using diagonals is due algorithm minimizing the number of convex pieces with Steiner points to Keil [Kei85]. The appears in [CD85]. Feng and Pavlidis [FP75b] give a heuristic algorithm for polygon decomposition and apply it to optical character recognition. Art gallery problems are an interesting topic related to polygon covering, where we seek to position the minimum number of guards in a given polygon such that every point in the interior of the polygon is watched by at least one guard. This corresponds to covering the polygon with a minimum number of starshaped polygons. O'Rourke [O'R87] is a beautiful book which presents the art gallery problem and its many variations. Related Problems: Triangulation (see page

), set cover (see page

).

Next: Simplifying Polygons Up: Computational Geometry Previous: Medial-Axis Transformation Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE194.HTM (3 of 3) [19/1/2003 1:31:53]

Simplifying Polygons

Next: Shape Similarity Up: Computational Geometry Previous: Polygon Partitioning

Simplifying Polygons

Input description: A polygon or polyhedron p, with n vertices. Problem description: Find a polygon or polyhedron p' with n' vertices, where the shape of p' is close to p and n' < n. Discussion: Polygon simplification has two primary applications. The first is in cleaning up a noisy representation of a polygon, perhaps obtained by scanning a picture of an object. By processing it, we hope to remove the noise and reconstruct the original object. The second is in data compression, where given a large and complicated object, we seek to simplify it by reducing detail. Ideally, we obtain a polygon with far fewer vertices that looks essentially the same. This can be a big win in computer graphics, where replacing a large model with a smaller model might have little visual impact but be significantly faster to render. Several issues arise in shape simplification: ●

Do you want the convex hull? - The simplest way to simplify a polygon is to take the convex hull ). The convex hull removes all internal concavities from the of its vertices (see Section polygon. If you were simplifying a robot model for motion planning, this is a good thing, because

file:///E|/BOOK/BOOK5/NODE195.HTM (1 of 4) [19/1/2003 1:31:55]

Simplifying Polygons

●

●

you are unlikely to be able to take advantage of the concavities in finding paths. If you were building an OCR system, the convex hull would be disastrous, because the concavities of characters provide most of the interesting features. An `X' would be identical to an `I', since both hulls are boxes. Another problem is that if the polygon is already convex, taking the convex hull will do nothing to simplify it further. Am I allowed to insert or just delete points? - What is typically needed is a way to represent the object as well as possible using only a given number of vertices. The simplest approaches employ local modifications to the boundary of the polygon, in order to reduce the vertex count. For example, if three consecutive vertices form a small-area triangle or define an extremely large angle, the center vertex can be deleted and replaced with an edge without severely distorting the polygon. Methods that only delete vertices quickly melt the shape into unrecognizability, however. More robust heuristics move vertices around to cover up the gaps that are created by deletions. Such ``split-and-merge'' heuristics can do a decent job, although nothing is guaranteed. Better results are likely by using the Douglas-Peucker algorithm, described below. Must the resulting polygon be intersection-free? - A serious drawback of such incremental procedures is that they fail to ensure simple polygons, those without self-intersections. Thus the ``simplified'' polygon may have artifacts that look ugly and that may cause problems for subsequent routines working on the polygon. If simplicity is important, you should test all the line segments of your polygon for any pairwise intersections, as discussed in Section

.

An approach to polygon simplification that guarantees a simple approximation involves computing minimum-link paths. The link distance of a path between points and t is the number of straight segments on the path. An as-the-crow-flies path has a link distance of one. In general, the link distance is one more than the number of turns. The link distance between points and t in a scene with obstacles is defined by the minimum link distance over all paths from to t. The link distance approach ``fattens'' the boundary of the polygon by some acceptable error

●

window (see Section ) in order to construct a channel around the polygon and then constructs the minimum-link path through this channel. The minimum-link cycle in this channel represents the simplest polygon that never deviates from the original boundary by more than . It constructs a globally optimal simplification that will not self-intersect, at the cost of implementation and time complexity. Are you given an image to clean up instead of a polygon to simplify? - The conventional, nongeometric approach to cleaning up noise from a digital image is to take the Fourier transform of the image, filter out the high-frequency elements, and then take the inverse transform to recreate the image. See Section

for details on the fast Fourier transform.

The Douglas-Plucker algorithm for shape simplification starts with a simple approximation and then refines it, instead of starting with a complicated polygon and trying to simplify it. Start by selecting two

file:///E|/BOOK/BOOK5/NODE195.HTM (2 of 4) [19/1/2003 1:31:55]

Simplifying Polygons

vertices and of polygon P, and propose the degenerate polygon as a simple approximation P'. Scan through each of the vertices of P, and select the one that is farthest from the corresponding edge of the polygon P'. Inserting this vertex adds the triangle to P' so as to minimize the maximum deviation from P. Points can be inserted until satisfactory results are achieved. This takes O(kn) to insert k points when |P|=n. Simplification becomes considerably more difficult in three dimensions. For example, it is NP-complete to find the minimum-size surface separating two polyhedra. Higher-dimensional analogies of the planar algorithms discussed here can be used to heuristically simplify polyhedra. See the references below. Implementations: A program for automatically generating level-of-detail hierarchies for polygonal models is available from http://www.cs.unc.edu/ geom/envelope.html and is free for noncommercial use. The user specifies a single error tolerance, and the maximum surface deviation of the simplified model from the original model, and a new, simplified model is generated. This code preserves holes and prevents self-intersection. Yet another approach to polygonal simplification is based on simplifying and expanding the medial-axis transform of the polygon. The medial-axis transform (see Section ) produces a skeleton of the polygon, which can be trimmed before inverting the transform to yield a simpler polygon. MAT [Ogn93] is a medial-axis transform code designed for 2D skeletonization and inversion of binary images, written by Robert Ogniewicz and available from http://hrl.harvard.edu/people/postdocs/rlo/rlo.dir/rlo-soft.html.

Notes: See [HG95] for a thorough survey of algorithms for shape simplification. It is also available from http://www.cs.cmu.edu/afs/cs/user/ph/www/heckbert.html, along with implementations. The Douglas-Peucker incremental refinement algorithm [DP73] is the basis for most shape simplification schemes, with a faster implementation due to Hershberger and Snoeyink [HS94]. The link distance approach to polygon simplification is presented in [GHMS93]. Shape simplification problems become considerably more complex in three dimensions. Even finding the minimum-vertex convex polyhedron lying between two nested convex polyhedra is NP-complete [DJ92], although approximation algorithms are known [MS95b]. Testing whether a polygon is simple can be performed in linear time, at least in theory, as a consequence of Chazelle's linear-time triangulation algorithm [Cha91]. Related Problems: Fourier transform (see page

file:///E|/BOOK/BOOK5/NODE195.HTM (3 of 4) [19/1/2003 1:31:55]

), convex hull (see page

).

Simplifying Polygons

Next: Shape Similarity Up: Computational Geometry Previous: Polygon Partitioning Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE195.HTM (4 of 4) [19/1/2003 1:31:55]

Shape Similarity

Next: Motion Planning Up: Computational Geometry Previous: Simplifying Polygons

Shape Similarity

Input description: Two polygonal shapes, Problem description: How similar are

and

and

.

?

Discussion: Shape similarity is a problem that underlies much of pattern recognition. Consider a system for optical character recognition (OCR). We have a known library of shape models representing letters and the unknown shapes we obtain by scanning a page. We seek to identify an unknown shape by matching it to the most similar shape model. The problem of shape similarity is inherently ill-defined, because what ``similar'' means is application dependent. Thus no single algorithmic approach can solve all shape matching problems. Whichever method you select, expect to spend a large chunk of time tweaking it so as to achieve maximum performance. Don't kid yourself - this is a difficult problem. Among your possible approaches are: ●

Hamming Distance - Assume that your two polygons have been overlaid one on top of the other. The Hamming distance measures the area of symmetric difference between the two polygons, in

file:///E|/BOOK/BOOK5/NODE196.HTM (1 of 3) [19/1/2003 1:31:56]

Shape Similarity

other words, the area of the regions lying within one of the two polygons but not both of them. If the two polygons are identical, and properly aligned, the Hamming distance will be zero. If the polygons differ only in a little noise at the boundary, then the Hamming distance will be small if the polygons are properly aligned. If the two polygons are completely disjoint, then the Hamming distance is the sum of the areas of the two polygons. Computing the area of the symmetric difference reduces to finding the intersection or union of , and then computing areas, as discussed in . The two polygons, as discussed in Section difficult part of computing Hamming distance is finding the right alignment, or overlay, of the two polygons. For certain applications, such as OCR, the overlay problem is simplified because the characters are inherently aligned with the page and thus not free to rotate. Hamming distance is particularly simple and efficient to compute on bit-mapped images, since after alignment all we do is sum the differences of the corresponding pixels.

●

●

Although Hamming distance makes sense conceptually and can be simple to implement, it captures only a crude notion of shape and is likely to be ineffective in most applications. Hausdorff distance - An alternative distance metric is Hausdorff distance, which identifies the point on that is the maximum distance from and returns this distance. The Hausdorff distance is not symmetrical, for the distance from to is not necessarily the distance from to . Note the difference between Hamming and Hausdorff distance. A large but thin protrusion from one of the models will have a large effect on the Hausdorff distance but little on the Hamming distance. However, a fattening of the entire boundary of one of the models (as is liable to happen with boundary noise) by a small amount will increase the Hamming distance yet have little effect on the Hausdorff distance. Which is better, Hamming or Hausdorff? It depends upon your application. As with Hamming distance, computing the right alignment between the polygons can be difficult and timeconsuming. Again, Hausdorff distance captures only a crude notion of shape. Comparing Skeletons - A more powerful approach to shape similarity uses thinning (see Section ) to extract a tree-like skeleton for each object. This skeleton captures many aspects of the original shape. The problem now reduces to comparing the shape of two such skeletons, using such features as the topology of the tree and the lengths/slopes of the edges. This comparison can

●

), with edges allowed to match be modeled as a form of subgraph isomorphism (see Section whenever their lengths and slopes are sufficiently similar. Neural Networks - A final method for shape comparison uses neural networks, which are . Neural nets prove a reasonable approach to recognition problems when discussed in Section you have a lot of data to experiment with and no particular ideas of what to do with it. First, you must identify a set of easily computed features of the shape, such as area, number of sides, and number of holes. Based on these features, a black-box program (the neural network training algorithm) takes your training data and produces a classification function. This classification function accepts as input the values of these features and returns a measure of what the shape is,

file:///E|/BOOK/BOOK5/NODE196.HTM (2 of 3) [19/1/2003 1:31:56]

Shape Similarity

or how close it is to a particular shape. How good are the resulting classifiers? It depends upon the application. Like any ad hoc method, neural networks usually take a fair amount of tweaking and tuning to realize their full potential. One caveat. Because your classifier was developed by a black box, you never really know why your classifier is making its decisions, so you can't know when it will fail. An interesting case was a system built for the military to distinguish between images of cars and tanks. It performed very well on test images but disastrously in the field. Eventually, someone realized that the car images had been filmed on a sunnier day than the tanks, and the program was classifying solely on the presence of clouds in the background of the image! Implementations: The Stuttgart Neural Network Simulator supports many types of networks and training algorithms, as well as sophisticated graphical visualization tools under X11. It has been ported to many flavors of UNIX. It is available for ftp from ftp.informatik.uni-stuttgart.de [129.69.211.2] in directory /pub/SNNS as SNNSv4.1.tar.gz (1.4 MB, Source code) and SNNSv4.1.Manual.ps.gz (1 MB, Documentation). It may be best to first have a look at the file SNNSv4.1.Readme. More information can be found in the WWW under http://www.informatik.uni-stuttgart.de/ipvr/bv/projekte/snns/snns.html An alternate distance metric between polygons can be based on its angle turning function [ACH An implementation in C of this turning function metric by Eugene K. Ressler is provided on the algorithm repository http://www.cs.sunysb.edu/ algorith.

91].

Notes: General books on pattern classification algorithms include [DH73, JD88]. A wide variety of computational geometry approaches to shape similarity testing have been proposed, including [AMWW88, ACH 91, Ata84, AE83, BM89, OW85]. A linear-time algorithm for computing the Hausdorff distance between two convex polygons is given in [Ata83], with algorithms for the general case reported in [HK90]. Related Problems: Graph isomorphism (see page

), thinning (see page

).

Next: Motion Planning Up: Computational Geometry Previous: Simplifying Polygons Algorithms Mon Jun 2 23:33:50 EDT 1997 file:///E|/BOOK/BOOK5/NODE196.HTM (3 of 3) [19/1/2003 1:31:56]

Motion Planning

Next: Maintaining Line Arrangements Up: Computational Geometry Previous: Shape Similarity

Motion Planning

Input description: A polygonal-shaped robot in a given starting position in a room containing polygonal obstacles, with a desired ending position t. Problem description: Find the shortest path in the room taking to t without intersecting any of the obstacles. Discussion: The difficulty of motion planning will be obvious to anyone who has ever had to move a large piece of furniture into a small apartment. The problem of motion planning also arises in systems for molecular docking. Many drugs are small molecules that act by binding to a given target model. The problem of identifying which binding sites are accessible to a candidate drug is clearly an instance of motion planning. Plotting paths for mobile robots is another canonical motion-planning application. Motion planning also provides a tool for computer animation. Given a set of object models that appear in two different scenes and , a motion planning algorithm can construct a short sequence of intermediate motions to transform to . These motions can serve to fill in the intermediate scenes between and , with such scene interpolation greatly reducing the amount of work the animator has to do.

file:///E|/BOOK/BOOK5/NODE197.HTM (1 of 4) [19/1/2003 1:31:58]

Motion Planning

There is a wide range in the complexity of motion planning problems, with many factors to consider: ●

Is your robot a point? - When the robot is a point, the problem becomes finding the shortest path from to t around the obstacles, also known as geometric shortest path. The most readily implementable approach constructs the visibility graph of the polygonal obstacles, plus the points and t. This visibility graph has a vertex for each obstacle vertex and an edge between two obstacle vertices if they ``see'' each other without being blocked by some obstacle edge. A brute-force algorithm to construct the visibility graph tests each candidate edge against the obstacle edges for a total time of time. By weighting each edge of the visibility graph with

●

●

●

●

its length and using Dijkstra's shortest-path algorithm (see Section ) on this graph, we can find the shortest path from to t in time bounded by the time to construct the graph. How is your robot free to move? - Motion planning becomes considerably more difficult when the robot becomes a polygon instead of a point. Now we must make sure that all of the corridors we use are wide enough to permit the robot to pass through. The complexity depends upon the number of degrees of freedom the robot has to move. Is it free to rotate as well as to translate? Does the robot have links that are free to bend or to rotate independently, as in an arm with a hand? Each degree of freedom corresponds to a dimension in the search space. Therefore, the more freedom, the harder it is to compute a path between two locations, although it becomes more likely that such a path exists. Can you simplify the shape of your robot? - Motion planning algorithms tend to be complex and time-consuming. Anything you can do to simplify your environment would be a win. In particular, consider replacing your robot by an enclosing disk. If there is a path for the disk, there will be a path for whatever is inside of it. Further, since any orientation of a disk is equivalent to any other orientation, rotation provides no help in finding a path Therefore, all movements are limited to translation only. Are motions limited to translation only? - When rotation is not allowed, the expanded obstacles approach can be used to reduce the problem to that of motion planning for a point robot, which is simply the shortest path in a graph. Pick a reference point on the robot, and replace each obstacle ). This creates a larger obstacle, by the Minkowski sum of the object and the robot (see Section defined as the robot walks a loop around the obstacle while maintaining contact with it. Finding a path from the initial position of the reference point to the goal point amidst these fattened obstacles defines a legal path for the polygonal robot. Are the obstacles known in advance? - Thus far we have assumed that the robot starts out with a map of its environment. This is not always true, or even possible, in applications where the obstacles move. There are two approaches to solving motion planning problems without a map. The first approach explores the environment, building a map of what has been seen, and then uses this map to plan a path to the goal. A simpler strategy, which will fail in environments of sufficient complexity, proceeds like a sightless man with a compass. Walk in the direction towards the goal until progress is blocked by an obstacle, and then trace a path along the obstacle until the robot is again free to proceed directly towards the goal.

file:///E|/BOOK/BOOK5/NODE197.HTM (2 of 4) [19/1/2003 1:31:58]

Motion Planning

The most practical approach to motion planning involves randomly sampling the configuration space of the robot. The configuration space defines the set of legal positions for the robot and has one dimension for each degree of freedom. For a planar robot capable of translation and rotation, the degrees of freedom are the x- and y-coordinates of a reference point on the robot and the angle relative to this point. Certain points in this space represent legal positions, while others intersect obstacles. Construct a set of legal configuration-space points by random sampling. For each pair of points and , decide whether there exists a direct, nonintersecting path between them. Construct a graph with vertices for each legal point and edges for each such traversable pair. The problem of finding a motion between two arbitrary positions reduces to seeing if there is a direct path from the initial/final position to some vertex in the graph, and then solving a shortest-path problem in the graph. There are lots of ways to enhance this basic technique for specific applications, such as adding additional vertices to regions of particular interest. This is a nice, clean approach for solving problems that would get very messy otherwise. Implementations: An implementation of collision detection (not really motion planning) is the I_COLLIDE collision detection library. For more information, check out the I_COLLIDE WWW page: http://www.cs.unc.edu/ geom/I_COLLIDE.html. O'Rourke [O'R94] gives a toy implementation of an algorithm to plot motion for a two-jointed robot arm in the plane. See Section

.

Notes: Motion planning was originally studied by Schwartz and Sharir as the ``piano mover's problem.'' Their solution constructs the complete free space of robot positions which do not intersect obstacles, and then finds the shortest path within the proper connected component. These free space descriptions are very complicated, involving arrangements of higher-degree algebraic surfaces. The fundamental papers on the piano mover's problem appear in [HSS87], with [SS90] being a survey of current results. The best general result for this free space approach to motion planning is due to Canny [Can87], who showed that , although faster algorithms exist for any problem with d degrees of freedom can be solved in special cases of the general motion planning problem. Latombe's book [Lat91] describes practical approaches to motion planning, including the random sampling method described above. The expanded obstacle approach to motion planning is due to LozanoPerez and Wesley [LPW79]. The heuristic, sightless man's approach to motion planning discussed above has been studied by Lumelski [LS87]. The time complexity of algorithms based on the free-space approach to motion planning depends intimately on the combinatorial complexity of the arrangement of surfaces defining the free space. Algorithms for maintaining arrangements are presented in Section

file:///E|/BOOK/BOOK5/NODE197.HTM (3 of 4) [19/1/2003 1:31:58]

. Davenport-Schintzl sequences

Motion Planning

often arise in the analysis of such arrangements. Sharir and Agarwal [SA95] provide a comprehensive treatment of Davenport-Schintzl sequences and their relevance to motion planning. Kedem and Sharir [KS90] give an time algorithm to find a path (not necessarily shortest) to algorithm for moving a line translate a convex k-gon from to t. Vegter [Veg90] gives an optimal segment (often called a ladder) in the plane with both translation and rotation. Related Problems: Shortest path (see page

), Minkowski sum (see page

).

Next: Maintaining Line Arrangements Up: Computational Geometry Previous: Shape Similarity Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE197.HTM (4 of 4) [19/1/2003 1:31:58]

Maintaining Line Arrangements

Next: Minkowski Sum Up: Computational Geometry Previous: Motion Planning

Maintaining Line Arrangements

Input description: A set of lines and line segments

.

Problem description: What is the decomposition of the plane defined by

?

Discussion: One of the most fundamental problems in computational geometry is constructing arrangements of lines, that is, explicitly building the regions formed by the intersections of a set of n lines. Algorithms for a surprising number of problems are based on constructing and analyzing the arrangement of a specific set of lines: ●

●

Degeneracy testing - Given a set of n lines in the plane, do any three of them pass through the same point? Brute-force testing of all triples takes time. Instead, we can construct the arrangement of the lines and then walk over each vertex and explicitly count its degree, all in quadratic time. Satisfying the maximum number of linear constraints - Suppose that we are given a set of n linear . Which point in the plane satisfies the largest number of constraints, each of the form them? Construct the arrangement of the lines. All points in any region or cell of this arrangement satisfy exactly the same set of constraints, so we need to test only one point per cell in order to find the global maximum.

file:///E|/BOOK/BOOK5/NODE198.HTM (1 of 4) [19/1/2003 1:32:00]

Maintaining Line Arrangements

Thinking of geometric problems in terms of the appropriate features in an arrangement can be very useful in formulating algorithms. Unfortunately, it must be admitted that arrangements are not as popular in practice as might be supposed. First, a certain depth of understanding is required to apply them correctly. Second, there have been few available implementations of the fundamental algorithms, a situation that is partially addressed below. ●

What do you want to do with the arrangement? - Given an arrangement and a query point, we are often interested in identifying which cell of the arrangement contains the point. This is the . Given an arrangement of lines or line problem of point location, discussed in Section segments, we are often interested in computing all points of intersection of the lines. The problem

●

●

. of intersection detection is discussed in Section How big will your arrangement be? - Algorithms for constructing arrangements are incremental. Beginning with an arrangement of one or two lines, subsequent lines are inserted into the arrangement one at a time, building larger and larger arrangements. To insert a new line, we start on the leftmost cell containing the line and walk over the arrangement to the right, moving from cell to neighboring cell and splitting into two pieces those cells that contain the new line. A geometric fact called the zone theorem implies that the kth line inserted cuts through k cells of the arrangement, and further that O(k) total edges form the boundary of these cells. This means that we can scan through each edge of every cell we encounter on our insertion walk, confident that linear total work will be performed while inserting the line into the arrangement. Therefore, the total time to insert all n lines in constructing the full arrangement is . Does your input consist of points instead of lines? - Although lines and points seem to be different geometric objects, such appearances can be misleading. Through the use of duality transformations, we can turn line L into point p and vice versa:

Duality is important because we can now apply line arrangements to point problems, often with surprising results. For example, suppose we are given a set of n points, and we want to know whether any three of them all lie on the same line. This sounds similar to the degeneracy testing problem discussed above. Not only is it similar, it is exactly the same, with only the role of points and lines exchanged. The answer follows from taking our points, dualizing them into lines as above, constructing the arrangement as above, and then searching for a vertex with three lines passing through it. The dual of this vertex gives the line on which the three initial vertices lie. Once we have constructed an arrangement through incremental methods, it often becomes useful to traverse each face of the arrangement exactly once. Such traversals are called sweepline algorithms and

file:///E|/BOOK/BOOK5/NODE198.HTM (2 of 4) [19/1/2003 1:32:00]

Maintaining Line Arrangements

are discussed in some detail in Section . The basic procedure is to sort the intersection points by xcoordinate and then walk from left to right while keeping track of all we have seen. Implementations: Arrange is a package written in C by Michael Goldwasser for maintaining arrangements of polygons in either the plane or on the sphere. Polygons may be degenerate and hence represent arrangements of lines. A randomized incremental construction algorithm is used and efficient point location on the arrangement supported. Polygons may be inserted but not deleted from the arrangement, and arrangements of several thousand vertices and edges can be constructed in a few seconds. Arrange is available from http://theory.stanford.edu/people/wass/wass.html. ) provides a function that constructs an embedded planar graph from a set of line LEDA (see Section segments, essentially constructing their arrangement. Notes: Edelsbrunner [Ede87] provides a comprehensive treatment of the combinatorial theory of arrangements, plus algorithms on arrangements with applications. It is an essential reference for anyone seriously interested in the subject. Good expositions on constructing arrangements include [O'R94]. Arrangements generalize naturally beyond two dimensions. Instead of lines, the space decomposition is defined by planes (or beyond 3-dimensions, hyperplanes). In general dimensions, the zone theorem states , and any single that any arrangement of n d-dimensional hyperplanes has total complexity . This provides the justification for the incremental hyperplane intersects cells of complexity construction algorithm for arrangements. Walking around the boundary of each cell to find the next cell that the hyperplane intersects takes time proportional to the number of cells created by inserting the hyperplane. The history of the zone theorem has become somewhat muddled, because the original proofs were later found to be in error in higher dimensions. See [ESS93] for a discussion and a correct proof. The theory of Davenport-Schintzl sequences is intimately tied into the study of arrangements. It is presented in [SA95]. The naive algorithm for sweeping an arrangement of lines sorts the intersection points by x-coordinate time. The topological sweep [EG89, EG91] eliminates the need to sort, and and hence requires it traverses the arrangement in quadratic time. This algorithm is readily implementable and can be applied to speed up many sweepline algorithms. Related Problems: Intersection detection (see page

file:///E|/BOOK/BOOK5/NODE198.HTM (3 of 4) [19/1/2003 1:32:00]

), point location (see page

).

Maintaining Line Arrangements

Next: Minkowski Sum Up: Computational Geometry Previous: Motion Planning Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE198.HTM (4 of 4) [19/1/2003 1:32:00]

Minkowski Sum

Next: Set and String Problems Up: Computational Geometry Previous: Maintaining Line Arrangements

Minkowski Sum

Input description: Point sets or polygons A and B, with n and m vertices, respectively. Problem description: What is the convolution of A and B, i.e. the Minkowski sum ? Discussion: Minkowski sums are useful geometric operations that can be used to fatten objects in appropriate ways. For example, a popular approach to motion planning for polygonal robots in a room ) fattens each of the obstacles by taking the Minkowski sum of with polygonal obstacles (see Section them with the shape of the robot. This reduces the problem to moving a point from the start to the goal ), using a standard shortest-path algorithm. Another application is in shape simplification (see Section where we fatten the boundary of an object to create a channel and then define as the shape the minimum link path lying within this channel. Similarly, convolving an irregular object with a small circle will help smooth out the boundaries by eliminating minor nicks and cuts. The definition of Minkowski sum assumes that the polygons A and B have been positioned on a coordinate system:

file:///E|/BOOK/BOOK5/NODE199.HTM (1 of 3) [19/1/2003 1:32:01]

Minkowski Sum

where x+y is the vector sum of two points. Thinking of this in terms of translation, the Minkowski sum is the union of all translations of A by a point defined within B. Issues arising in computing Minkowski sums include: ●

●

Are your objects rasterized images or explicit polygons? - The definition of Minkowski summation suggests a simple algorithm if A and B are rasterized images, and thus contain a number of pixels proportional to their area. Initialize a sufficiently large matrix of pixels by determining the size of the convolution of the bounding boxes of A and B. For each pair of points in A and B, sum up their coordinates and darken the appropriate pixel. These algorithms get somewhat more complicated if an explicit polygonal representation of the Minkowski sum is needed. Are your objects convex or nonconvex? - The complexity of computing Minkowski sum depends in a serious way on the shape of the polygons. If both A and B are convex, the Minkowski sum can be found in O(n+m) time by tracing the boundary of one polygon with another. If one of them is nonconvex, the size of the sum alone can be as large as . Even worse is when both A and B are nonconvex, in which case the size of the sum can be as large as . Be aware that the Minkowski sum of nonconvex polygons can have a certain ugliness to it. For example, holes can be either created or destroyed.

Although more efficient algorithms exist, a straightforward approach to computing the Minkowski sum is based on triangulation and union. First, triangulate both polygons, then compute the Minkowski sum of each triangle of A against each triangle of B. The sum of a triangle against another triangle is easy to compute and is a special case of convex polygons, discussed below. The union of these O(n m) convex polygons will be A+B. Algorithms for computing the union of polygons are based on plane sweep, as discussed in Section

.

Computing the Minkowski sum of two convex polygons is easier than the general case, because the sum will always be convex. For convex polygons, it is easiest to slide A along the boundary of B and compute the sum edge by edge. This is the best approach for triangles against triangles as well. Implementations: To date, we have not uncovered a suitable Minkowski sum code. When we do, it will be made available on http://www.cs.sunysb.edu/ algorith, the Algorithm Repository site. Notes: Good expositions on algorithms for Minkowski sums include [O'R94]. The fastest algorithms for various cases of Minkowski sums include [KOS91, Sha87]. Kedem and Sharir [KS90] present an efficient algorithm for translational motion planning for polygonal robots, based on Minkowski sums. Related Problems: Thinning (see page

), motion planning (see page

file:///E|/BOOK/BOOK5/NODE199.HTM (2 of 3) [19/1/2003 1:32:01]

), simplifying polygons (see

Minkowski Sum

page

).

Next: Set and String Problems Up: Computational Geometry Previous: Maintaining Line Arrangements Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE199.HTM (3 of 3) [19/1/2003 1:32:01]

Set and String Problems

Next: Set Cover Up: A Catalog of Algorithmic Previous: Minkowski Sum

Set and String Problems Sets and strings both represent collections of objects. The primary difference is whether order matters. Sets are collections of symbols whose order is assumed to carry no significance, while the arrangement of symbols is exactly what defines a string. The assumption of a canonical order makes it possible to solve string problems much more efficiently than set problems, through techniques such as dynamic programming and advanced data structures like suffix trees. The interest in and importance of string processing algorithms have been increasing, largely due to biological and text-processing applications. A product of this interest are three recent books on string algorithms: ●

●

●

● ● ● ● ● ● ● ●

Crochemore and Rytter [CR94] - A comprehensive book on advanced string algorithms, but somewhat formal and fairly difficult to follow. Stephen [Ste94] - A reasonably gentle introduction to basic string algorithmics. Possibly the best available book for the beginner. Gusfield [Gus97] - This is now the most comprehesive introduction to string algorithms. It contains a thorough discussion on suffix trees, with new, clear formulations of classical exact string-matching algorithms.

Set Cover Set Packing String Matching Approximate String Matching Text Compression Cryptography Finite State Machine Minimization Longest Common Substring

file:///E|/BOOK/BOOK5/NODE200.HTM (1 of 2) [19/1/2003 1:32:02]

Set and String Problems ●

Shortest Common Superstring

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE200.HTM (2 of 2) [19/1/2003 1:32:02]

Set Cover

Next: Set Packing Up: Set and String Problems Previous: Set and String Problems

Set Cover

Input description: A set of subsets

of the universal set

Problem description: What is the smallest subset T of S such that

. ?

Discussion: Set cover arises when you try to efficiently acquire or represent items that have been packaged in a fixed set of lots. You want to obtain all the items, while buying as few lots as possible. Finding a cover is easy, because you can always buy one of each lot. However, by finding a small set cover you can do the same job for less money. An interesting application of set cover is Boolean logic minimization. We are given a particular Boolean function of k variables, which for each of the possible input vectors describes whether the desired output is 0 or 1. We seek the simplest circuit that exactly implements this function. One approach is to find a disjunctive normal form (DNF) formula on the variables and their complements, such as . We could build one ``and'' term for each input vector and then ``or'' them all together, but we might save considerably by factoring out common subsets of variables. Given a set of feasible ``and'' terms, each of which covers a subset of the vectors we need, we seek to ``or'' together the smallest number of terms that realize the function. This is exactly the set cover problem.

file:///E|/BOOK/BOOK5/NODE201.HTM (1 of 4) [19/1/2003 1:32:04]

Set Cover

There are several variations of set cover problems to be aware of: ●

●

Are you allowed to cover any element more than once? - The distinction here is between set . If we are allowed to covering and set packing, the latter of which is discussed in Section cover elements more than once, as in the logic minimization problem above, we should take advantage of this freedom, as it usually results in a smaller covering. Are your sets derived from the edges or vertices of a graph? - Set cover is a very general problem, and it includes several useful graph problems as special cases. Suppose instead that you seek the smallest set of edges in a graph that covers each vertex at least once. The solution is to find a ), and then add arbitrary edges to cover the maximum matching in the graph (see Section remaining vertices. Suppose you seek the smallest set of vertices in a graph that covers each edge at least once. This is the vertex cover problem, discussed in Section

●

●

.

It is instructive to model vertex cover as an instance of set cover. Let the universal set U be the set of edges . Construct n subsets, with consisting of the edges incident on vertex . Although vertex cover is just a set cover problem in disguise, you should take advantage of the fact that better algorithms exist for vertex cover. Do your subsets contain only two elements each? - When all of your subsets have at most two elements each, you are in luck. This is about the only special case that you can solve efficiently to optimality, by using the matching technique described above. Unfortunately, as soon as your subsets get to have three elements each, the problem becomes NP-complete. Do you want to cover elements with sets, or sets with elements? - In the hitting set problem, we seek a small number of items that together represent an entire population. Formally, we are given a set of subsets of the universal set , and we seek the smallest such that each subset contains at least one element of T. Thus for all subset . Suppose we desire a small Congress with at least one representative of each ethnic group. If each ethnic group is represented as a subset of people, the minimum hitting set is the smallest possible politically correct Congress. Hitting set is illustrated in Figure

file:///E|/BOOK/BOOK5/NODE201.HTM (2 of 4) [19/1/2003 1:32:04]

.

Set Cover

Figure: Hitting set is dual to set cover Hitting set is, in fact, dual to set cover, meaning it is exactly the same problem in disguise. Replace each element of U by a set of the names of the subsets that contain it. Now S and U have exchanged roles, for we seek a set of subsets from U to cover all the elements of S. This is is exactly the set cover problem. Thus we can use any of the set cover codes below to solve hitting set after performing this simple translation. Since the vertex cover problem is NP-complete, the set cover problem must be at least as hard. In fact, it is somewhat harder. Approximation algorithms do no worse than twice optimal for vertex cover, but only a times optimal approximation algorithm exists for set cover. The greedy heuristic is the right approach for set cover. Begin by placing the largest subset in the set cover, and then mark all its elements as covered. We will repeatedly add the subset containing the largest number of uncovered elements, until all elements are covered. This heuristic always gives a set cover times as many sets as optimal, and in practice it usually does a lot better. using at most The simplest implementation of the greedy heuristic sweeps through the entire input instance of m subsets for each greedy step. However, by using such data structures as linked lists and a bounded-height ), the greedy heuristic can be implemented in O(S) time, where priority queue (see Section is the size of the input representation. It pays to check whether there are certain elements that exist in only a few subsets, ideally only one. If so, we should select the biggest subsets containing these elements at the very beginning. We will have to take them eventually, and they carry with them extra elements that we might have to pay to cover by waiting until later. Simulated annealing is likely to produce somewhat better set covers than these simple heuristics, if that is important for your application. Backtracking can be used to guarantee you an optimal solution, but typically it is not worth the computational expense. Implementations: Pascal implementations of an exhaustive search algorithm for set packing, as well as heuristics for set cover, appear in [SDK83]. See Section

.

Notes: Survey articles on set cover include [BP76]. See [CK75] for a computational study of set cover algorithms. An excellent exposition on algorithms and reduction rules for set cover is presented in [SDK83]. Good expositions of the greedy heuristic for set cover include [CLR90]. An example demonstrating that the greedy heuristic for set cover can be as bad as is presented in [Joh74, PS82]. This is not a defect file:///E|/BOOK/BOOK5/NODE201.HTM (3 of 4) [19/1/2003 1:32:04]

Set Cover

of the heuristic. Indeed, it is provably hard to approximate set cover to within an approximation factor better than [LY93]. Related Problems: Matching (see page

), vertex cover (see page

), set packing (see page

Next: Set Packing Up: Set and String Problems Previous: Set and String Problems Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE201.HTM (4 of 4) [19/1/2003 1:32:04]

).

Set Packing

Next: String Matching Up: Set and String Problems Previous: Set Cover

Set Packing

Input description: A set of subsets

of the universal set

.

Problem description: What is the largest number of mutually disjoint subsets from S? Discussion: Set packing problems arise in partitioning applications, where we need to partition elements under strong constraints on what is an allowable partition. The key feature of packing problems is that no elements are permitted to be covered by more than one set. Consider the problem of finding the maximum independent set in a graph G, discussed in Section . We seek a large subset of vertices such that each edge is adjacent to at most one of the selected vertices. To model this as set packing, let the universal set consist of all edges of G, and subset consist of all edges incident on vertex . Any set packing corresponds to a set of vertices with no edge in common, in other words, an independent set. Scheduling airline flight crews to airplanes is another application of set packing. Each airplane in the fleet needs to have a crew assigned to it, consisting of a pilot, copilot, and navigator. There are constraints on the composition of possible crews, based on their training to fly different types of aircraft, as well as any personality conflicts. Given all possible crew and plane combinations, each represented by a subset of items, we need an assignment such that each plane and each person is in exactly one chosen file:///E|/BOOK/BOOK5/NODE202.HTM (1 of 3) [19/1/2003 1:32:06]

Set Packing

combination. After all, the same person cannot be on two different planes, and every plane needs a crew. We need a perfect packing given the subset constraints. Set packing is used here to represent a bunch of problems on sets, all of which are NP-complete and all of which are quite similar: ●

Must every element from the universal set appear in one selected subset? - In the exact cover problem, we seek some collection of subsets such that each element is covered exactly once. The airplane scheduling problem above has the flavor of exact covering, since every plane and crew has to be employed. Unfortunately, exact cover is similar to that of Hamiltonian cycle in graphs. If we really must cover all the elements exactly once, and this existential problem is NP-complete, then all we can do is exponential search. This will be prohibitive unless there are so many solutions that we will stumble upon one quickly. Things will be far better if we can be content with a partial solution, say by adding each element of U as a singleton subset of S. Thus we can expand any set packing into an exact cover by mopping up the unpacked elements of U with singleton sets. Now our problem is reduced to finding a minimum-cardinality set packing, which can be attacked via heuristics, as discussed below.

●

What is the penalty for covering elements twice? - In set cover (see Section ), there is no penalty for elements existing in many selected subsets. In pure set packing, any such violation is forbidden. For many such problems, the truth lies somewhere in between. Such problems can be approached by charging the greedy heuristic more to select a subset that contains previously covered elements than one that does not.

The right heuristics for set packing are greedy, and similar to those of set cover (see Section ). If we seek a packing with many sets, then we repeatedly select the smallest subset, delete all subsets from S that clash with it, and repeat. If we seek a packing with few subsets, we do the same but always pick the largest possible subset. As usual, augmenting this approach with some exhaustive search or randomization (in the form of simulated annealing) is likely to yield better packings at the cost of additional computation. Implementations: Since set cover is a more popular and more tractable problem than set packing, it might be easier to find an appropriate implementation to solve the cover problem. Many such implementations should be readily modifiable to support certain packing constraints. Pascal implementations of an exhaustive search algorithm for set packing, as well as heuristics for set cover, appear in [SDK83]. See Section

for details on ftp-ing these codes.

file:///E|/BOOK/BOOK5/NODE202.HTM (2 of 3) [19/1/2003 1:32:06]

Set Packing

Notes: An excellent exposition on algorithms and reduction rules for set packing is presented in [SDK83], including the airplane scheduling application discussed above. Survey articles on set packing include [BP76]. Related Problems: Independent set (see page

), set cover (see page

Next: String Matching Up: Set and String Problems Previous: Set Cover Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE202.HTM (3 of 3) [19/1/2003 1:32:06]

).

String Matching

Next: Approximate String Matching Up: Set and String Problems Previous: Set Packing

String Matching

Input description: A text string t of length n. A pattern string p of length m. Problem description: Find the first (or all) instances of the pattern in the text. Discussion: String matching is fundamental to database and text processing applications. Every text editor must contain a mechanism to search the current document for arbitrary strings. Pattern matching programming languages such as Perl and Awk derive much of their power from their built-in string matching primitives, making it easy to fashion programs that filter and modify text. Spelling checkers scan an input text for words in the dictionary and reject any strings that do not match. Several issues arise in identifying the right string matching algorithm for the job: ●

Are your search patterns and/or texts short? - If your strings are sufficiently short and your queries sufficiently infrequent, the brute-force O(m n)-time search algorithm will likely suffice. For each possible starting position , it tests whether the m characters starting from the ith position of the text are identical to the pattern. For very short patterns, say

, you can't hope to beat brute force by much, if at all, and you

file:///E|/BOOK/BOOK5/NODE203.HTM (1 of 4) [19/1/2003 1:32:08]

String Matching

●

●

●

shouldn't try. Further, since we can reject the possibility of a match at a given starting position the instant we observe a text/pattern mismatch, we expect much better than O(mn) behavior for typical strings. Indeed, the trivial algorithm usually runs in linear time. But the worst case certainly can occur, as with pattern and text . What about longer texts and patterns? - By being more clever, string matching can be performed in worst-case linear time. Observe that we need not begin the search from scratch on finding a character mismatch between the pattern and text. We know something about the subsequent characters of the string as a result of the partial match from the previous position. Given a long partial match from position i, we can jump ahead to the first character position in the pattern/text that will provide new information about the text starting from position i+1. The Knuth-MorisPratt algorithm preprocesses the search pattern to construct such a jump table efficiently. The details are tricky to get correct, but the resulting implementations yield short, simple programs. Even better in practice is the Boyer-Moore algorithm, although it offers similar worst-case performance. Boyer-Moore matches the pattern against the text from right to left, in order to avoid looking at large chunks of text on a mismatch. Suppose the pattern is abracadabra, and the eleventh character of the text is x. This pattern cannot match in any of the first eleven starting positions of the text, and so the next necessary position to test is the 22nd character. If we get very lucky, only n/m characters need ever be tested. The Boyer-Moore algorithm involves two sets of jump tables in the case of a mismatch: one based on pattern matched so far, the other on the text character seen in the mismatch. Although somewhat more complicated than Knuth-Morris-Pratt, it is worth it in practice for patterns of length m > 5. Will you perform multiple queries on the same text? - Suppose you were building a program to repeatedly search a particular text database, such as the Bible. Since the text remains fixed, it pays build a data structure to speed up search queries. The suffix tree and suffix array data structures, , are the right tools for the job. discussed in Section Will you search many texts using the same patterns? - Suppose you were building a program to screen out dirty words from a text stream. Here, the set of patterns remains stable, while the search texts are free to change. In such applications, we may need to find all occurrences of each of k different patterns, where k can be quite large. Performing a linear-time search for each of these patterns yields an O(k (m+n)) algorithm. If k is large, a better solution builds a single finite automaton that recognizes each of these patterns and returns to the appropriate start state on any character mismatch. The Aho-Corasick algorithm builds such an automaton in linear time. Space savings can be achieved by optimizing the pattern recognition automaton, as discussed in Section version of fgrep.

. This algorithm was used in the original

Sometimes multiple patterns are specified not as a list of strings, but concisely as a regular expression. For example, the regular expression matches any string on (a,b,c) that begins and ends with an a, including a itself. The best way to test whether input strings are described by a given regular expression R is to construct the finite automaton equivalent to R and file:///E|/BOOK/BOOK5/NODE203.HTM (2 of 4) [19/1/2003 1:32:08]

String Matching

then simulate the machine on the string. Again, see Section automata from regular expressions.

●

for details on constructing

When the patterns are specified by context-free grammars instead of regular expressions, the problem becomes one of parsing, discussed in books on compilers and programming languages. What if our text or pattern contains a spelling error? - Finally, observe that the algorithms discussed here work only for exact string matching. If you must allow some tolerance for spelling errors, your problem is approximate string matching, which is throughly discussed in Section

.

Implementations: SPARE Parts is a string pattern recognition toolkit, written in C++ by Bruce Watson. It provides production-quality implementations of all major variants of the classical string matching algorithms for single patterns (both Knuth-Morris-Pratt and Boyer-Moore) and multiple patterns (both Aho-Corasick and Commentz-Walter). SPARE Parts is available by anonymous ftp from ftp.win.tue.nl in /pub/techreports/pi/watson.phd/. A greatly improved commercial version is available from www.RibbitSoft.com. ) provides animations for both the Boyer-Moore and Knuth-Morris-Pratt XTango (see Section algorithms. The C source code for each animation is included. Implementations in C and Pascal of several algorithms for exact and approximate string matching appear in [GBY91]. Sedgewick provides similar implementations of Knuth-Morris-Pratt, Rabin-Karp, and Boyer-Moore in C++. See Section

for details on both codes.

Implementations in C of the Boyer-Moore, Aho-Corasick, and regular expression matching algorithms appear in [BR95]. The code for these algorithms is printed in the text and available on disk for a modest fee. Notes: All books on string algorithms contain thorough discussions of exact string matching, including [CR94, Ste94, Gus97]. Good expositions on the Boyer-Moore [BM77] and Knuth-Morris-Pratt algorithms [KMP77] include [Baa88, CLR90, Man89]. Aho [Aho90] provides a good survey on algorithms for pattern matching in strings, particularly where the patterns are regular expressions instead of strings, and for the Aho-Corasick algorithm for multiple patterns [AC75]. An algorithm merging Aho-Corasick and Boyer-Moore can be faster for small numbers of patterns [CW79], but the window where it wins is apparently fairly narrow. Empirical comparisons of string matching algorithms include [DB86, Hor80, dVS82]. Which algorithm performs best depends upon the properties of the strings and the size of the alphabet. For long patterns file:///E|/BOOK/BOOK5/NODE203.HTM (3 of 4) [19/1/2003 1:32:08]

String Matching

and texts, I recommend that you use the best implementation of Boyer-Moore that you can find. An interesting classical problem is determining the minimum number of comparisons needed to perform exact string matching. A version of Boyer-Moore never makes more than 2n comparisons independent of the number of occurrences of the pattern in the text [AG86]. More recent results are very technical, depending upon the details of the model and the alphabet size. There is a lower bound of n-m+1 text characters that any algorithm must be examine in the worst case [Riv77]. The history of string matching algorithms is somewhat checkered because several published proofs were incorrect or incomplete [Gus97]. The Karp-Rabin algorithm [KR87] uses a hash function to perform string matching in linear expected time. Its worst-case time remains quadratic, and its performance in practice appears somewhat worse than the character comparison methods described above. Related Problems: Suffix trees (see page

), approximate string matching (see page

Next: Approximate String Matching Up: Set and String Problems Previous: Set Packing Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE203.HTM (4 of 4) [19/1/2003 1:32:08]

).

Approximate String Matching

Next: Text Compression Up: Set and String Problems Previous: String Matching

Approximate String Matching

Input description: A text string t and a pattern string p. An edit cost bound k. Problem description: Can we transform t to p using at most k insertions, deletions, and substitutions? Discussion: Approximate string matching is fundamental to text processing, because we live in an errorprone world. Any spelling correction program must be able to identify the closest match for any text string not found in a dictionary. Genbank has become a fundamental tool for molecular biology by supporting homology (similarity) searches on DNA sequences. Suppose you were to sequence a new gene in man, and you discovered that it is similar to the hemoglobin gene in rats. It is likely that this new gene also produces hemoglobin, and any differences are the result of genetic mutations during evolution.

I once encountered approximate string matching in evaluating the performance of an optical character recognition system that we built. After scanning and recognizing a test document, we needed to compare the correct answers with those produced by our system. To improve our system, it was important to count how often each pair of letters were getting confused and to identify gibberish when the program was trying to make out letters where none existed. The solution was to do an alignment between the two texts. Insertions and deletions corresponded to gibberish, while substitutions signaled errors in our recognizers.

file:///E|/BOOK/BOOK5/NODE204.HTM (1 of 5) [19/1/2003 1:32:10]

Approximate String Matching

This same principle is used in file difference programs, which identify the lines that have changed between two versions of a file. When no errors are permitted, the problem becomes that of exact string matching, which is presented in Section

. Here, we restrict our discussion to dealing with errors.

Dynamic programming provides the basic approach to approximate string matching, as discussed in Section . Let D[i,j] denote the cost of editing the first i characters of the text string t into the first j characters of the pattern string p. The recurrence follows because we must have done something with and . The only options are matching them, substituting one for the other, deleting , or inserting a match for . Thus D[i,j] is the minimum of the costs of these possibilities: 1. If then D[i-1,j-1] else D[i-1,j-1] + substitution cost. 2. D[i-1,j] + deletion cost of . 3. D[i,j-1] + deletion cost of . Several issues remain before we can make full use of this recurrence: ●

●

Do I want to match the pattern against the full text, or against a substring? - Appropriately initializing the boundary conditions of the recurrence distinguishes between the algorithms for string matching and substring matching. Suppose we want to align the full text against the full pattern. Then the cost of D[i,0] must be that of deleting the first i characters of the text, so D[i,0] = i. Similarly, D[0,j] = j. Now suppose that the pattern may occur anywhere within the text. The proper cost of D[i,0] is now 0, since there should be no penalty for starting the alignment in the ith position. The cost of D[0,j] remains j, however, since the only way to match the first j pattern characters with nothing is to delete all of them. The cost of the best pattern match against the text will be given by . How should I select the substitution and insertion/deletion costs? - The basic algorithm can be easily modified to use different costs for insertion, deletion, and the substitutions of specific pairs of characters. Which costs you use depends upon what you are planning to do with the alignment.

The most common cost assignment charges the same for insertion, deletion, or substitution. Charging a substitution cost of more than insertion + deletion is a good way to ensure that substitutions never get performed, since it will be cheaper to edit both characters out of the string. With just insertion and deletion to work with, the problem reduces to longest common subsequence, discussed in Section . Often, it pays to tweak the edit distance costs and study the resulting alignments until you find a set that does the job.

file:///E|/BOOK/BOOK5/NODE204.HTM (2 of 5) [19/1/2003 1:32:10]

Approximate String Matching ●

●

●

●

How do I get the actual alignment of the strings, instead of just the cost? - To obtain the transcript of the editing operations performed in achieving the minimum cost alignment, we can work backwards from the complete cost matrix D. To get to cell D[n,m], we had to come from one of D[n-1,m] (insertion), D[n,m-1] (deletion), or D[n-1,m-1] (substitution/match). Which of the three options was chosen can be reconstructed given these costs and the characters and . By continuing to work backwards from the previous cell, we can trace the entire path and thus reconstruct the alignment. What if the two strings are very similar to each other? - The dynamic programming algorithm can be thought of as finding a shortest path across an grid, where the cost of each edge depends upon which operation it represents. If we seek an alignment involving a combination of at most d insertions, deletions, and substitutions, we need only traverse the band of O(d n) cells within a distance d of the central diagonal. If no such low-cost alignment exists within this band, then no low-cost alignment can exist in the full cost matrix. How can I minimize the required storage? - Dynamic programming takes both quadratic time and space. For many applications, the space required to store the dynamic programming table is a space is needed to compute D[m,n], much more serious problem. Observe that only since we need only maintain two active rows (or columns) of the matrix in order to compute the final value. The entire matrix is required only if we need to reconstruct the actual sequence alignment. To save space, we can use Hirschberg's clever recursive algorithm. During one pass of the linearspace algorithm above to compute D[m,n], we maintain all the values for the (m/2)nd column and identify which middle-element cell D[m/2,x] was used to optimize D[m,n]. This reduces our problem to finding the best paths from D[1,1] to D[m/2,x] and from D[m/2,x] to D[m/2,n], both of which can be solved recursively. Each time we throw away half of the matrix elements from consideration, and so the total time remains O(mn). This linear-space algorithm proves to be a big win in practice on long strings, although it is somewhat more difficult to program. Does string similarity mean that the strings sound alike? - Other models of approximate pattern matching become more appropriate than edit distance in certain applications. Particularly interesting is Soundex, a hashing scheme that attempts to pair up English words that sound alike. This can be useful in testing whether two names that have been spelled differently are likely to be the same. For example, my last name is often spelled ``Skina'', ``Skinnia'', ``Schiena'', and occasionally ``Skiena''. All of these hash to the same Soundex code, S25. The algorithm works by dropping vowels and silent letters, removing doubled letters, and then assigning the remaining letters numbers from the following classes: BFPV gets a 1, CGJKQSXZ gets a 2, DT gets a 3, L gets a 4, MN gets a 5, and R gets a 6. The code starts with the first letter and contains at most three digits. Although this all sounds very hokey, experience shows that it works reasonably well. Experience indeed: Soundex has been used since the 1920's.

Implementations: The best available software tools for approximate pattern matching are glimpse and agrep [WM92a, WM92b], developed by Manber and Wu at the University of Arizona and available from

file:///E|/BOOK/BOOK5/NODE204.HTM (3 of 5) [19/1/2003 1:32:10]

Approximate String Matching

http://glimpse.cs.arizona.edu/. Glimpse is a tool for building and using an index to search through file systems, while agrep (approximate general regular expression pattern matcher) is a tool supporting text search with spelling errors. Both programs are widely used and respected. ht://Dig is an alternative WWW text search engine from Andrew Scherpbier, which contains implementations of Soundex and Metaphone. It is available from http://htdig.sdsu.edu/ and is released under the GNU general public license. Implementations in C of the Soundex and dynamic programming edit-distance algorithms appear in [BR95]. The code for these algorithms is printed in the text and is available on disk for a modest fee. Bare bones implementations in C and Pascal of several algorithms for exact and approximate string matching appear in [GBY91]. See Section

for further details.

Notes: The wide range of applications for approximate string matching was made apparent in Sankoff and Kruskal's book [SK83], which remains a useful historical reference for the problem. Surveys on approximate pattern matching include [HD80]. The basic dynamic programming alignment algorithm is attributed to [WF74], although it is apparently folklore. The edit distance between two strings is sometimes referred to as the Levenshtein distance. Expositions of dynamic programming to compute Levenshtein distance include [Baa88, CLR90, Man89]. Expositions of Hirschberg's linear-space algorithm [Hir75] include [CR94, Gus97]. Masek and Paterson [MP80] compute the edit distance between m- and n-length strings in time for constant-sized alphabets, using ideas from the four Russians algorithm for Boolean matrix multiplication [ADKF70]. The shortest path formulation leads to a variety of algorithms algorithm due to Myers [Mye86]. that are good when the edit distance is small, including an time [HS77], as presented in [Man89]. Longest increasing subsequence can be done in Soundex was invented and patented by M. K. Odell and R. C. Russell. Expositions on Soundex include [BR95, Knu73b]. Metaphone is a recent attempt to improve on Soundex [BR95, Par90]. Related Problems: String matching (see page

), longest common substring (see page

Next: Text Compression Up: Set and String Problems Previous: String Matching

file:///E|/BOOK/BOOK5/NODE204.HTM (4 of 5) [19/1/2003 1:32:10]

).

Approximate String Matching

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE204.HTM (5 of 5) [19/1/2003 1:32:10]

Text Compression

Next: Cryptography Up: Set and String Problems Previous: Approximate String Matching

Text Compression

Input description: A text string S. Problem description: A shorter text string S' such that S can be correctly reconstructed from S'. Discussion: Secondary storage devices fill up quickly on every computer system, even though their capacity doubles each year. Decreasing storage prices have only increased interest in data compression, since there is now more data to compress than ever before. Data compression is the algorithmic problem of finding alternative, space-efficient encodings for a given data file. With the rise of computer networks, a new mission for data compression has arisen, that of increasing the effective bandwidth of networks by reducing the number of bits before transmission. Data compression is a problem for which practical people like to invent ad hoc methods, designed for their particular applications. Sometimes these outperform general methods, but often they do not. The following issues arise in selecting the right data compression algorithm: ●

Must we exactly reconstruct the input text after compression? - A primary issue in data compression algorithms is the question of lossy versus lossless encodings. Text applications typically demand lossless encodings, because users become disturbed whenever their data files

file:///E|/BOOK/BOOK5/NODE205.HTM (1 of 4) [19/1/2003 1:32:11]

Text Compression

●

●

●

●

have been corrupted. However, fidelity is not such an issue in image or video compression, where the presence of small artifacts will be imperceptible to the viewer. Significantly greater compression ratios can be obtained using lossy compression, which is why all image/video/audio compression algorithms take advantage of this freedom. Can I simplify my data before I compress it? - The most effective way to free up space on a disk is to delete files you don't need. Likewise, any preprocessing you can do to a file to reduce its information content before compression will pay off later in better performance. For example, is it possible to eliminate extra blank spaces or lines from the file? Can the document be converted entirely to uppercase characters or have formatting information removed? Does it matter whether the algorithm is patented? - One concern is that many data compression algorithms are patented, in particular the LZW variation of the Lempel-Ziv algorithm discussed below. Further, Unisys, the owner of the patent, makes periodic attempts to collect. My personal (although not legal) recommendation is to ignore them, unless you are in the business of selling text compression software. If this makes you uncomfortable, note that there are other variations on the Lempel-Ziv algorithm that are not under patent protection and perform about as well. See the notes and implementations below. How do I compress image data - Run-length coding is the simplest lossless compression algorithm for image data, where we replace runs of identical pixel values with one instance of the pixel and an integer giving the length of the run. This works well on binary images with large regions of similar pixels (like scanned text) and terribly on images with many quantization levels and a little noise. It can also be applied to text with many fields that have been padded by blanks. Issues like how many bits to allocate to the count field and the traversal order converting the twodimensional image to a stream of pixels can have a surprisingly large impact on the compression ratio. For serious image and video compression applications, I recommend that you use a lossy coding method and not fool around with implementing it yourself. JPEG is the standard highperformance image compression method, while MPEG is designed to exploit the frame-to-frame coherence of video. Encoders and decoders for both are provided in the implementation section. Must compression and decompression both run in real time? - For many applications, fast decompression is more important than fast compression, and algorithms such as JPEG exist to take advantage of this. While compressing video for a CD-ROM, the compression will be done only once, while decompression will be necessary anytime anyone plays it. In contrast, operating systems that increase the effective capacity of disks by automatically compressing each file will need a symmetric algorithm with fast compression times as well.

Although there are literally dozens of text compression algorithms available, they are characterized by two basic approaches. In static algorithms, such as Huffman codes, a single coding table is built by analyzing the entire document. In adaptive algorithms, such as Lempel-Ziv, a coding table is built on the fly and adapts to the local character distribution of the document. An adaptive algorithm will likely prove to be the correct answer: ●

Huffman codes - Huffman codes work by replacing each alphabet symbol by a variable-length

file:///E|/BOOK/BOOK5/NODE205.HTM (2 of 4) [19/1/2003 1:32:11]

Text Compression

code string. ASCII uses eight bits per symbol in English text, which is wasteful, since certain characters (such as `e') occur far more often than others (such as `q'). Huffman codes compress text by assigning `e' a short code word and `q' a longer one. Optimal Huffman codes can be constructed using an efficient greedy algorithm. Sort the symbols in increasing order by frequency. We will merge the two least frequently used symbols x and y into a new symbol m, whose frequency is the sum of the frequencies of its two child symbols. By replacing x and y by m, we now have a smaller set of symbols, and we can repeat this operation n1 times until all symbols have been merged. Each merging operation defines a node in a binary tree, and the left or right choices on the path from root-to-leaf define the bit of the binary code word for each symbol. Maintaining the list of symbols sorted by frequency can be done using priority queues, which yields an -time Huffman code construction algorithm.

●

Although they are widely used, Huffman codes have three primary disadvantages. First, you must make two passes over the document on encoding, the first to gather statistics and build the coding table and the second to actually encode the document. Second, you must explicitly store the coding table with the document in order to reconstruct it, which eats into your space savings on short documents. Finally, Huffman codes exploit only nonuniformity in symbol distribution, while adaptive algorithms can recognize the higher-order redundancy in strings such as 0101010101.... Lempel-Ziv algorithms - Lempel-Ziv algorithms, including the popular LZW variant, compress text by building the coding table on the fly as we read the document. The coding table available for compression changes at each position in the text. A clever protocol between the encoding program and the decoding program ensures that both sides of the channel are always working with the exact same code table, so no information can be lost. Lempel-Ziv algorithms build coding tables of recently-used text strings, which can get arbitrarily long. Thus it can exploit frequently-used syllables, words, and even phrases to build better encodings. Further, since the coding table alters with position, it adapts to local changes in the text distribution, which is important because most documents exhibit significant locality of reference. The truly amazing thing about the Lempel-Ziv algorithm is how robust it is on different types of files. Even when you know that the text you are compressing comes from a special restricted vocabulary or is all lowercase, it is very difficult to beat Lempel-Ziv by using an applicationspecific algorithm. My recommendation is not to try. If there are obvious application-specific redundancies that can safely be eliminated with a simple preprocessing step, go ahead and do it. But don't waste much time fooling around. No matter how hard you work, you are unlikely to get significantly better text compression than with gzip or compress, and you might well do worse.

Implementations: A complete list of available compression programs is provided in the comp.compression FAQ (frequently asked questions) file, discussed below. This FAQ will likely point you to what you are looking for, if you don't find it in this section. file:///E|/BOOK/BOOK5/NODE205.HTM (3 of 4) [19/1/2003 1:32:11]

Text Compression

The best general-purpose program for text compression is gzip, which implements a public domain variation of the Lempel-Ziv algorithm. It is distributed under the GNU software licence and can by obtained from ftp://prep.ai.mit.edu/pub/gnu/gzip-1.2.4.tar. Unix compress is another popular compression program based on the patented LZW algorithm. It is available from ftp://wuarchive.wustl.edu/packages/compression/compress-4.1.tar. A JPEG implementation is available from ftp://ftp.uu.net/graphics/jpeg/jpegsrc.v6a.tar.gz. MPEG can be found at ftp://havefun.stanford.edu/pub/mpeg/MPEGv1.2.2.tar.Z. Algorithm 673 [Vit89] of the Collected Algorithms of the ACM is a Pascal implementation of dynamic Huffman codes, which is a one-pass, adaptive text compression algorithm. See Section fetching this program.

for details on

Notes: Many books on data compression are available, but we highly recoomend Bell, Cleary, and Witten [BCW90] and Storer [Sto88]. Another good source of information is the USENET newsgroup comp.compression. Check out its particularly comprehensive FAQ (frequently asked questions) compendium at location ftp://rtfm.mit.edu/pub/usenet/news.answers/compression-faq. Good expositions on Huffman codes [Huf52] include [AHU83, BR95, CLR90, Eve79a, Man89]. Expositions on the LZW [Wel84, ZL78] algorithm include [BR95]. There is an annual IEEE Data Compression Conference, the proceedings of which should be studied seriously before attempting to develop a new data compression algorithm. On reading the proceedings, it will become apparent that this is a mature technical area, where much of the current work (especially for text compression) is shooting for fairly marginal improvements on special applications. On a more encouraging note, we remark that this conference is held annually at a world-class ski resort in Utah. Related Problems: Shortest common superstring (see page

), cryptography (see page

Next: Cryptography Up: Set and String Problems Previous: Approximate String Matching Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE205.HTM (4 of 4) [19/1/2003 1:32:11]

).

Cryptography

Next: Finite State Machine Minimization Up: Set and String Problems Previous: Text Compression

Cryptography

Input description: A plaintext message T or encrypted text E, and a key k. Problem description: Encode T using k giving E, or decode E using k back to T. Discussion: Cryptography has grown substantially in importance in recent years, as computer networks have made confidential documents more vulnerable to prying eyes. Cryptography is a way to increase security by making messages difficult to read if they fall into the wrong hands. Although the discipline of cryptography is at least two thousand years old, its algorithmic and mathematical foundations have recently solidified to the point where there can now be talk of provably secure cryptosystems. There are three classes of cryptosystems everyone should be aware of: ●

Caesar shifts The oldest ciphers involve mapping each character of the alphabet to a different letter. The weakest such ciphers rotate the alphabet by some fixed number of characters (often 13), and thus have only 26 possible keys. Better is to use an arbitrary permutation of the letters, so there are 26! possible keys. Even so, such systems can be easily attacked by counting the frequency of each symbol and exploiting the fact that `e' occurs more often than `z'. While there are variants that will make this more difficult to break, none will be as secure as DES or

file:///E|/BOOK/BOOK5/NODE206.HTM (1 of 5) [19/1/2003 1:32:13]

Cryptography

●

●

RSA. Data Encryption Standard (DES) - This algorithm is based on repeatedly shuffling the bits of your text as governed by the key. The standard key length for DES (56 bits) is now considered too short for applications requiring the highest level of security. However, a simple variant called triple DES permits an effective key length of 112 bits by using three rounds of DES with two 56bit keys. In particular, first encrypt with key1, then decrypt with key2, before finally encrypting with key1. There is a mathematical reason for using three rounds instead of two, and the encryptdecrypt-encrypt pattern is used so that the scheme is equivalent to single DES when key1 = key2. Rivest-Shamir-Adelman (RSA) - RSA is a public key cryptosystem, meaning that different keys are used to encode and decode messages. Since the encoding key is of no help in decoding, it can be made public at no risk to security. The security of RSA is based on the difference in the ). Encoding is computational complexity of factoring and primality testing (see Section (relatively) fast because it relies on primality testing to construct the key, while the hardness of decryption follows from that of factoring. Still, RSA is slow relative to other cryptosystems, roughly 100 to 1,000 times slower than DES.

The key issue in selecting a cryptosystem is identifying your paranoia level, i.e. deciding how much security you need. Who are you trying to stop from reading your stuff: your grandmother, local thieves, the Mafia, or the NSA? If you can use an accepted implementation of RSA, such as PGP discussed below, you should feel safe against just about anybody. If there is an implementation of DES on your machine, that will likely be good enough for most applications. For example, I use DES to encrypt my final exam each semester, and it proved more than sufficient the time an ambitious student broke into my office looking for it. If the NSA had been breaking in, the story might have been different, although even here it is important to understand that the most serious security holes are human, not algorithmic. Making sure your password is long enough, hard to guess, and not written down is far more important than obsessing about the encryption algorithm. Simple ciphers like the Caesar shift are fun and easy to program. For this reason, it is perhaps healthy to use them for applications needing only a casual level of security (such as hiding the punchlines of jokes). Since they are easy to break, they should never be used for serious security applications. One thing that you should never do is mess around with developing your own novel cryptosystem. The security of DES and RSA is accepted largely because these systems have both survived over twenty years of public scrutiny. Over this period, many other cryptosystems have been proposed, proven vulnerable to attack, and then abandoned. This is not a field for amateurs. If you are charged with implementing a cryptosystem, carefully study a respected program such as PGP (discussed below) to see how its author, Philip Zimmermann, skillfully handled such issues as key selection and key distribution. A cryptosystem is as strong as its weakest link. There are several problems related to cryptography that arise often in practice:

file:///E|/BOOK/BOOK5/NODE206.HTM (2 of 5) [19/1/2003 1:32:13]

Cryptography ●

How can I validate the integrity of data against corruption? - In any communications application, there is need to validate that the transmitted data is identical to that which has been received. One solution is for the received to transmit the data back to the source and have the original sender confirm that the two texts are identical. This fails when the exact inverse of an error is made in the retransmission, but a more serious problem is that your available bandwidth is cut in half with such a scheme. A more efficient if less reliable method is to use a checksum, a simple mathematical function that hashes a long text down to a simple number or digit, and then transmit the checksum along with the text. The checksum can be recomputed on the receiving end and bells set off if the computed checksum is not identical to what was received. Perhaps the simplest checksum scheme just adds up the byte or character values and takes the sum modulo some constant, say . Unfortunately, an error transposing two or more characters would go undetected under such a scheme, since addition is commutative.

●

●

Cyclic-redundancy check (CRC) provides a more powerful method for computing checksums and is used in most communications systems and internally in computers to validate disk drive transfers. These codes compute the remainder in the ratio of two polynomials, the numerator of which is a function of the input text. The design of these polynomials involves considerable mathematical sophistication and ensures that all reasonable errors are detected. The details of efficient computation are complicated enough that we recommend that you start from an existing implementation, described below. How can I prove that a file has not been changed? - If I send you a contract in electronic form, what is to stop you from editing the file and then claiming that your version was what we had really agreed to? I need a way to prove that any modification to a document is fraudulent. Digital signatures are a cryptographic way for me to stamp my document as genuine. Given a file, I can compute a checksum for it, and then encrypt this checksum using my own private key. I send you the file and the encrypted checksum. You can now edit the file, but to fool the judge you must also edit the encrypted checksum such that it can be decrypted to the correct checksum. With a suitably good checksum function, designing a file that yields the same checksum becomes an insurmountable problem. How can I prove that I am me? - Authentication is the process of one party convincing another that they are who they say they are. The historical solutions have involved passwords or keys, so I prove that I am who I say I am because I know my credit card number or carry an ID card. The problem with such schemes is that anyone who eavesdrops on this conversation or who steals my physical key can now successfully impersonate me. What we need is some way for me to convince you that I know my key, without actually telling you the key. One such method to do so is for you to send me a random number or text, and I use my key to encrypt your text and send it back to you. If you then decrypt this text and compare it to the message you sent me, you gain confidence that I am who I say I am. We can repeat this exercise on several random texts until sufficient authentication has been agreed upon. If we use a

file:///E|/BOOK/BOOK5/NODE206.HTM (3 of 5) [19/1/2003 1:32:13]

Cryptography

secure enough cryptosystem, we can be confident that an eavesdropper will not be able to deduce my key even given several plain and encrypted texts. Such authentication protocols of back-and-forth messages often involve the use of randomness to frustrate eavesdroppers. Different protocols satisfy particular needs and constraints about who has to know what. It is important to do some reading before attempting to design your own protocols. References are provided below. Implementations: The USENET FAQ (frequently asked questions) file on cryptography provides a wealth of information, including pointers to implementations. Check it out at ftp://rtfm.mit.edu/pub/usenet/news.answers/cryptography-faq/. Distributing cryptographic software is complicated by United States export restrictions, which make it illegal to export encryption software. PGP (Pretty Good Privacy) is such a good implementation of RSA that its author Philip Zimmerman was charged with export violations by federal authorities. PGP may be obtained from the Electronic Frontier Foundation (EFF) at http://www.eff.org/pub/Net_info/Tools/Crypto/PGP/. A good discussion on checksums and cyclic-redundancy codes, with implementations in C, appear in [BR95]. The code for these algorithms is printed in the text and is available on disk for a modest fee. ) uses checksums to ensure that data files remain unmodified The Stanford Graphbase (see Section from the original distribution. Algorithm 536 [Kno79] of the Collected Algorithms of the ACM is an encryption function for passwords, written in Fortran. See Section

for further information.

Notes: Kahn [Kah67] presents the fascinating history of cryptography from ancient times to 1967 and is particularly noteworthy in light of the secretive nature of the subject. More recent and more technical works on cryptography include Denning [Den82] and Schneier [Sch94], the latter of which provides a through overview of different cryptographic algorithms, including implementations for sale. Rawlins [Raw92] provides a good introduction to cryptographic algorithms, from Caesar shift to public key to zero-knowledge proofs. An algorithm for breaking simple substitution ciphers appears in [PR79]. Expositions on the RSA algorithm [RSA78] include [CLR90]. The RSA Laboratories home page http://www.rsa.com/rsalabs/ is very informative. See [Sta95] for an excellent guide to PGP and its underlying algorithms. The history of DES is well presented in [Sch94]. Particularly controversial was the decision by the NSA to limit key length to 56 bits, presumably short enough to be cracked by special-purpose computers costing on the order of several million dollars. Despite some theoretical progress in breaking DES analytically [BS93], the most significant threat remains special-purpose hardware. file:///E|/BOOK/BOOK5/NODE206.HTM (4 of 5) [19/1/2003 1:32:13]

Cryptography

MD5 [Riv92] is the secure hashing function used by PGP to compute digital signatures. include [Sch94, Sta95]. Related Problems: Factoring and primality testing (see page

Expositions

), Text compression (see page

)).

Next: Finite State Machine Minimization Up: Set and String Problems Previous: Text Compression Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE206.HTM (5 of 5) [19/1/2003 1:32:13]

Finite State Machine Minimization

Next: Longest Common Substring Up: Set and String Problems Previous: Cryptography

Finite State Machine Minimization

Input description: A deterministic finite automaton M. Problem description: The smallest deterministic finite automaton M' such that M' behaves identically to M. Discussion: Problems associated with constructing and minimizing finite state machines arise repeatedly in software and hardware design applications. Finite state machines are best thought of as pattern recognizers, and minimum-size machines correspond to recognizers that require less time and space. Complicated control systems and compilers are often built using finite state machines to encode the current state and associated actions, as well as the set of possible transitions to other states. Minimizing the size of this machine minimizes its cost. Finite state machines are best thought of as edge-labeled directed graphs, where each vertex represents one of n states and each edge a transition from one state to the other on receipt of the alphabet symbol that labels the edge. The automaton above analyzes a given sequence of coin tosses, with the dark states signifying that an even number of heads have been observed. Automata can be represented using any graph data structure (see Section alphabet.

), or by using an

file:///E|/BOOK/BOOK5/NODE207.HTM (1 of 4) [19/1/2003 1:32:15]

transition matrix, where

is the size of the

Finite State Machine Minimization

Finite state machines are often used to specify search patterns in the guise of regular expressions, which are patterns formed by and-ing, or-ing, and looping over smaller regular expressions. For example, the regular expression matches any string on (a,b,c) that begins and ends with an a (including a itself). The best way to test whether a string is described by a given regular expression (especially if many strings will be tested) is to construct the equivalent finite automaton and then simulate the machine on the string. See Section

for alternative approaches to string matching.

We consider three different problems on finite automata: ●

Minimizing deterministic finite state machines - Transition matrices for finite automata quickly become prohibitively large for sophisticated machines, thus fueling the need for tighter encodings. The most direct approach is to eliminate redundant states in the automaton. As the example above illustrates, automata of widely varying sizes can compute the same function. Algorithms for minimizing the number of states in a deterministic finite automaton (DFA) appear in any book on automata theory. The basic approach works by partitioning the states into gross equivalence classes and then refining the partition. Initially, the states are partitioned into accepting, rejecting, and other classes. The transitions from each node branch to a given class on a given symbol. Whenever two states , t in the same class C branch to elements of different classes, the class C must be partitioned into two subclasses, one containing , the other containing t.

●

This algorithm makes a sweep though all the classes looking for a new partition, and repeating the process from scratch if it finds one. This yields an algorithm, since at most n-1 sweeps need ever be performed. The final equivalence classes correspond to the states in the minimum algorithm is known with available implementations automaton. In fact, a more efficient, discussed below. Constructing deterministic machines from nondeterministic machines - DFAs are simple to understand and work with, because the machine is always in exactly one state at any given time. Nondeterministic automata (NFAs) can be in more than one state at a time, so its current ``state'' represents a subset of all possible machine states. In fact, any NFA can be mechanically converted to an equivalent DFA, which can then be minimized as above. However, converting an NFA to a DFA can cause an exponential blowup in the number of states, which perversely might later be eliminated in the minimization. This exponential blowup makes automaton minimization problems NP-hard whenever you do not start with a DFA. The proofs of equivalence between NFAs, DFAs, and regular expressions are elementary enough to be covered in undergraduate automata theory classes. However, they are surprisingly nasty to implement, for reasons including but not limited to the exponential blowup of states. Implementations are discussed below.

file:///E|/BOOK/BOOK5/NODE207.HTM (2 of 4) [19/1/2003 1:32:15]

Finite State Machine Minimization ●

Constructing machines from regular expressions - There are two approaches to converting a regular expression to an equivalent finite automaton, the difference being whether the output automaton is to be a nondeterministic or deterministic machine. The former is easier to construct but less efficient to simulate. The nondeterministic construction uses -moves, which are optional transitions that require no input to fire. On reaching a state with an -move, we must assume that the machine can be in either state. Using such -moves, it is straightforward to construct an automaton from a depth-first traversal of the parse tree of the regular expression. This machine will have O(m) states, if m is the length of the regular expression. Further, simulating this machine on a string of length n takes O(m n) time, since we need consider each state/prefix pair only once. The deterministic construction starts with the parse tree for the regular expression, observing that each leaf represents one of the alphabet symbols in the pattern. After recognizing a prefix of the text, we can be left in some subset of these possible positions, which would correspond to a state in the finite automaton. The derivatives method builds up this automaton state by state as it is needed. Even so, some regular expressions of length m require states in any DFA implementing them, such as . There is no way to avoid this exponential blowup in the space required. Note, however, that it takes linear time to simulate an input string on any automaton, regardless of the size of the automaton.

Implementations: FIRE Engine is a finite automaton toolkit, written in C++ by Bruce Watson. It provides production-quality implementations of finite automata and regular expression algorithms. Several finite automaton minimization algorithms have been implemented, including Hopcroft's algorithm. Both deterministic and nondeterministic automata are supported. FIRE Engine has been used for compiler construction, hardware modeling, and computational biology applications. It is strictly a computing engine and does not provide a graphical user interface. FIRE Engine is available by anonymous ftp from ftp.win.tue.nl in the directory /pub/techreports/pi/watson.phd/. A greatly improved commercial version is available from www.RibbitSoft.com. Grail is a C++ package for symbolic computation with finite automata and regular expressions, from Darrell Raymond and Derrick Wood. Grail enables one to convert between different machine representations and to minimize automata. It can handle machines with 100,000 states and dictionaries of 20,000 words. All code and documentation are accessible from the WWW site http://www.csd.uwo.ca/research/grail, as well as pointers to a variety of other automaton packages. Commercial use of Grail is not allowed without approval, although it is freely available to students and educators. An implementation in C of a regular-expression matching algorithm appears in [BR95]. The source code for this program is printed in the text and is available on disk for a modest fee. A bare bones implementation in C of a regular-expression pattern matching algorithm appears in [GBY91]. See

file:///E|/BOOK/BOOK5/NODE207.HTM (3 of 4) [19/1/2003 1:32:15]

Finite State Machine Minimization

Section

.

XTango (see Section ) includes a simulation of a DFA. Many of the other animations (but not this one) are interesting and quite informative to watch. FLAP (Formal Languages and Automata Package) is a tool by Susan Rodger for drawing and simulating finite automata, pushdown automata, and Turing machines. Using FLAP, one can draw a graphical representation (transition diagram) of an automaton, edit it, and simulate the automaton on some input. FLAP was developed in C++ for X-Windows. See http://www.cs.duke.edu:80/ rodger/tools/tools.html.

Notes: Aho [Aho90] provides a good survey on algorithms for pattern matching, and a particularly clear exposition for the case where the patterns are regular expressions. The technique for regular expression pattern matching with -moves is due to Thompson [Tho68]. Other expositions on finite automaton pattern matching include [AHU74]. Hopcroft [Hop71] gave an optimal algorithm for minimizing the number of states in DFAs. The derivatives method of constructing a finite state machine from a regular expression is due to Brzozowski [Brz64] and has been expanded upon in [BS86]. Expositions on the derivatives method includes Conway []. Testing the equivalence of two nondeterministic finite state machines is PSPACE-complete [SM73]. Related Problems: Satisfiability (see page

). string matching (see page

).

Next: Longest Common Substring Up: Set and String Problems Previous: Cryptography Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE207.HTM (4 of 4) [19/1/2003 1:32:15]

Longest Common Substring

Next: Shortest Common Superstring Up: Set and String Problems Previous: Finite State Machine Minimization

Longest Common Substring

Input description: A set S of strings

.

Problem description: What is the longest string S' such that for each appear as a subsequence of ?

,

, the characters of S

Discussion: The problem of longest common subsequence arises whenever we search for similarities across multiple texts. A particularly important application is in finding a consensus among DNA sequences. The genes for building particular proteins evolve with time, but the functional regions must remain consistent in order to work correctly. By finding the longest common subsequence of the same gene in different species, we learn what has been conserved over time. The longest common substring problem is a special case of edit distance (see Section ), when substitutions are forbidden and only exact character match, insert, and delete are allowable edit operations. Under these conditions, the edit distance between p and t is n+m-2 |lcs(p,t)|, since we can delete the missing characters from p to the lcs(p,t) and insert the missing characters from t to transform p to t. This is particularly interesting because the longest common subsequence can be faster to compute

file:///E|/BOOK/BOOK5/NODE208.HTM (1 of 4) [19/1/2003 1:32:18]

Longest Common Substring

than edit distance. Issues arising include: ●

Are you looking for a common substring or scattered subsequence? - In detecting plagiarism or attempting to identify the authors of anonymous works, we might need to find the longest phrases shared between several documents. Since phrases are strings of consecutive characters, we need the longest common substring between the texts. The longest common substring of a set of strings can be identified in linear time using suffix trees, . The trick is to build a suffix tree containing all the strings, label each discussed in Section leaf with the set of strings that contain it, and then do a depth-first traversal to identify the deepest node that has descendents from each input string. For the rest of this discussion, we will restrict attention to finding common scattered subsequences. Dynamic programming can be used to find the longest common subsequence of two strings, S and T, of n and m characters each. This algorithm is a special case of the edit-

●

. Let M[i,j] denote the number of characters in the longest distance computation of Section common substring of and . In general, if , there is no way the . If S[i] = T[j], we have last pair of characters could match, so the option to select this character for our substring, so . This gives a recurrence that computes M, and thus finds the length of the longest common subsequence in O(nm) time. We can reconstruct the actual common substring by walking backward from M[n,m] and establishing which characters were actually matched along the way. What if there are relatively few sets of matching characters? - For strings that do not contain too many copies of the same character, there is a faster algorithm. Let r be the number of pairs of . Thus r can be as large as mn if both strings consist entirely of the positions (i,j) such that . This technique treats the same character, but r = n if the two strings are permutations of pairs of r as defining points in the plane. The complete set of r such points can be found in O(n + m + r) time by bucketing techniques. For each string, we create a bucket for each letter of the alphabet and then partition all of its characters into the appropriate buckets. For each letter c of the alphabet, create a point (,t) from every pair and , where and are buckets for c. A common substring represents a path through these points that moves only up and to the right, never down or to the left. Given these points, the longest such path can be found in time. We will sort the points in order of increasing x-coordinate (breaking ties in favor of increasing y-coordinate. We will insert these points one by one in this order, and for each k,

file:///E|/BOOK/BOOK5/NODE208.HTM (2 of 4) [19/1/2003 1:32:18]

Longest Common Substring

●

●

, maintain the minimum y-coordinate of any path going through exactly k points. Inserting a new point will change exactly one of these paths by reducing the y-coordinate of the path whose last point is barely greater than the new point. What if the strings are permutations? - If the strings are permutations, then there are exactly n time. A particularly pairs of matching characters, and the above algorithm runs in important case of this occurs in finding the longest increasing subsequence of a sequence of numbers. Sorting this sequence and then replacing each number by its rank in the total order gives us a permutation p. The longest common subsequence of p and gives the longest increasing subsequence. What if we have more than two strings to align? - The basic dynamic programming algorithm can time, where n is the length of the longest string. This be generalized to k strings, taking algorithm is exponential in the number of strings k, and so it will likely be too expensive for more than 3 to 4 strings. Further, the problem is NP-complete, so no better exact algorithm is destined to come along. This problem of multiple sequence alignment has received considerable attention, and numerous heuristics have been proposed. Many heuristics begin by computing the pairwise alignment between each of the pairs of strings, and then work to merge these alignments. One approach is to build a graph with a vertex for each character of each string. There will be an edge between and if the corresponding characters are matched in the alignment between S and T. Any kclique (see Section ) in this graph describes a commonly aligned character, and all such cliques can be found efficiently because of the sparse structure of this graph. Although these cliques will define a common subsequence, there is no reason to believe that it will be the longest such substring. Appropriately weakening the clique requirement provides a way to increase it, but still there can be no promises.

Implementations: MAP (Multiple Alignment Program) [Hua94] by Xiaoqiu Huang is a C language program that computes a global multiple alignment of sequences using an iterative pairwise method. Certain parameters will need to be tweaked to make it accommodate non-DNA data. It is available by anonymous ftp from cs.mtu.edu in the pub/huang directory. Combinatorica [Ski90] provides a Mathematica implementation of an algorithm to construct the longest increasing subsequence of a permutation, which is a special case of longest common subsequence. This algorithm is based on Young tableaux rather than dynamic programming. See Section

.

Notes: Good expositions on longest common subsequence include [AHU83, CLR90]. A survey of algorithmic results appears in [GBY91]. The algorithm for the case where all the characters in each sequence are distinct or infrequent is due to Hunt and Szymanski [HS77]. Expositions of this algorithm

file:///E|/BOOK/BOOK5/NODE208.HTM (3 of 4) [19/1/2003 1:32:18]

Longest Common Substring

include [Aho90, Man89]. Multiple sequence alignment for computational biology is treated in [Wat95]. Certain problems on strings become easier when we assume a constant-sized alphabet. Masek and Paterson [MP80] solve longest common subsequence in for constant-sized alphabets, using the four Russians technique. Related Problems: Approximate string matching (see page

), shortest common superstring (see page

).

Next: Shortest Common Superstring Up: Set and String Problems Previous: Finite State Machine Minimization Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE208.HTM (4 of 4) [19/1/2003 1:32:18]

Shortest Common Superstring

Next: Algorithmic Resources Up: Set and String Problems Previous: Longest Common Substring

Shortest Common Superstring

Input description: A set of strings

.

Problem description: Find the shortest string S' that contains each element of S as a substring. Discussion: Shortest common superstring arises in a variety of applications, including sparse matrix compression. Suppose we have an matrix with most of the elements being zero. We can partition each row into m/k runs of k elements each and construct the shortest common superstring S' of these runs. array of pointers into the We now have reduced the problem to storing the superstring, plus an superstring denoting where each of the runs starts. Accessing a particular element M[i,j] still takes constant time, but there is a space savings when |S| << mn. Another application arises in DNA sequencing. It happens to be easy to sequence small fragments of DNA, say up to about 500 base pairs or characters. However, the real interest is in sequencing large molecules. The standard approach to large-scale, ``shotgun'' sequencing clones many copies of the target molecule, breaks them randomly into small fragments, sequences the fragments, and then proposes the shortest superstring of the fragments as the correct sequence. While it is an article of faith that the shortest superstring will be the most likely sequence, this seems to work reasonably well in practice.

file:///E|/BOOK/BOOK5/NODE209.HTM (1 of 3) [19/1/2003 1:32:19]

Shortest Common Superstring

Finding a superstring of all the substrings is not difficult, as we can simply concatenate them all together. It is finding the shortest such string that is problematic. Indeed, shortest common superstring remains NPcomplete under all reasonable classes of strings. The problem of finding the shortest common superstring can easily be reduced to that of the traveling ). Create an overlap graph G where vertex represents string . Edge salesman problem (see Section will have weight equal to the length of minus the overlap of with . The path visiting all the vertices of minimum total weight defines the shortest common superstring. The edge weights of this and is not the same as the overlap of graph are not symmetric, after all, the overlap of and . Thus only programs capable of solving asymmetric TSPs can be applied to this problem. The greedy heuristic is the standard approach to approximating the shortest common superstring. Find the pair of strings with the maximum number of characters of overlap. Replace them by the merged string, and repeat until only one string remains. Given the overlap graph above, this heuristic can be efficiently ) and then merging if the implemented by inserting all of the edge weights into a heap (see Section appropriate ends of the two strings have not yet be used, which can be maintained with an array of Boolean flags. The potentially time-consuming part of this heuristic is in building the overlap graph. The brute-force approach to finding the maximum overlap of two length-l strings takes , which must be repeated times. Faster times are possible by appropriately using suffix trees (see Section ). Build a tree containing all suffixes of all reversed strings of S. String overlaps with if a suffix of matches a suffix of the reverse of . The longest overlap for each fragment can be found in time linear in its length. How well does the greedy heuristic perform? If we are unlucky with the input, the greedy heuristic can be fooled into creating a superstring that is at least twice as long as optimal. Usually, it will be a lot better in practice. It is known that the resulting superstring can never be more than 2.75 times optimal. Building superstrings becomes more difficult with positive and negative substrings, where negative substrings cannot be substrings of the superstring. The problem of deciding whether any such consistent substring exists is NP-complete, unless you are allowed to add an extra character to the alphabet to use as a spacer. Implementations: CAP (Contig Assembly Program) [Hua92] by Xiaoqiu Huang is a C language program supporting DNA shotgun sequencing by finding the shortest common superstring of a set of fragments. As to performance, CAP took 4 hours to assemble 1,015 fragments of a total of 252,000 characters on a Sun SPARCstation SLC. Certain parameters will need to be tweaked to make it accommodate non-DNA data. It is available by anonymous ftp from cs.mtu.edu in the pub/huang directory. file:///E|/BOOK/BOOK5/NODE209.HTM (2 of 3) [19/1/2003 1:32:19]

Shortest Common Superstring

Notes: The shortest common superstring problem and its application to DNA shotgun assembly is ably surveyed in [Wat95]. Kececioglu and Myers [KM95] report on an algorithm for the more general version of shortest common superstring, where the strings may have errors. Their paper is recommended reading to anyone interested in fragment assembly. Blum et al. [BJL 91] gave the first constant-factor approximation algorithms for shortest common superstring, with a variation of the greedy heuristic. More recent research has beaten the constant down to 2.75, progress towards the expected factor-two result. Related Problems: Suffix trees (see page

), text compression (see page

).

Next: Algorithmic Resources Up: Set and String Problems Previous: Longest Common Substring Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE209.HTM (3 of 3) [19/1/2003 1:32:19]

Software systems

Next: LEDA Up: Algorithmic Resources Previous: Algorithmic Resources

Software systems In this section, we describe several particularly comprehensive implementations of combinatorial algorithms, all of which are available over the Internet. Although these codes are discussed in the relevant sections of the catalog, they are substantial enough to warrant further attention. A good algorithm designer does not reinvent the wheel, and a good programmer does not rewrite code that other people have written. Picasso put it best: ``Good artists borrow. Great artists steal.'' However, a word of caution about stealing. Many of the codes described below (and throughout this book) have been made available for research or educational use, although commercial use requires a licensing arrangement with the author. I urge you to respect this. Licensing terms from academic institutions are usually surprisingly modest. The recognition that industry is using a particular code is important to the authors, often more important than the money involved. This can lead to enhanced support or future releases of the software. Do the right thing and get a license. Information about terms or whom to contact is usually available embedded within the documentation, or available at the source ftp/WWW site. Although the bulk of the systems we describe here are available by accessing our algorithm repository, http://www.cs.sunysb.edu/ algorith (as well as on the enclosed CD-ROM), we encourage you to get them from the original sites instead of Stony Brook. There are three reasons. First, the version on the original site is much more likely to be maintained. Second, there are often supporting files and documentation that we for whatever reason did not download, and which may be of interest to you. Finally, by ftp-ing from the original sites, you will keep traffic down at the algorithm repository site, which will minimize the complaints from our system staff that I anticipate if this service becomes very popular.

● ●

LEDA Netlib

file:///E|/BOOK/BOOK5/NODE211.HTM (1 of 2) [19/1/2003 1:32:20]

Software systems

Collected Algorithms of the ACM The Stanford GraphBase Combinatorica Algorithm Animations with XTango Programs from Books ❍ Discrete Optimization Algorithms in Pascal ❍ Handbook of Data Structures and Algorithms ❍ Combinatorial Algorithms for Computers and Calculators ❍ Algorithms from P to NP ❍ Computational Geometry in C ❍ Algorithms in C++ ❍

● ● ● ●

Next: LEDA Up: Algorithmic Resources Previous: Algorithmic Resources Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE211.HTM (2 of 2) [19/1/2003 1:32:20]

LEDA

Next: Netlib Up: Software systems Previous: Software systems

LEDA

LEDA, for Library of Efficient Data types and Algorithms, is perhaps the best single resource available to support combinatorial computing. It has been under development since 1988 by a group at MaxPlanck-Instutut in Saarbrücken, Germany, including Kurt Mehlhorn, Stefan Näher, Stefan Schirra, Christian Uhrig, and Christoph Burnikel. LEDA is unique because of (1) the algorithmic sophistication of its developers and (2) the level of continuity and resources invested in the project. LEDA is implemented in C++ using templates, and it should compile on most new compilers, but not some old ones. LEDA is available by anonymous ftp from ftp.mpi-sb.mpg.de in directory /pub/LEDA, or at http://www.mpi-sb.mpg.de/LEDA/leda.html. The distribution contains all sources, installation instructions, and a substantial users manual [NU95]. An active Usenet newsgroup comp.lang.c++.leda is inhabited by users of LEDA. A good article on LEDA is available [MN95], and a book is promised soon. LEDA is not in the public domain, but it can be used freely for research and teaching. Commerical licenses are also available. What LEDA offers is a complete collection of well-implemented data structures and types. Particularly useful is the graph type, which supports all the basic operations one needs in an intelligent way, although this generality comes at some cost in size and speed over handcrafted implementations. A small but useful library of graph algorithms is included, which illustrates how cleanly and concisely these algorithms can be implemented using the LEDA data types. Good implementations of the most important data structures supporting such common data types as dictionaries and priority queues are provided. There are also algorithms and data structures for computational geometry in the plane, including some support for X-windows. LEDA Implementation

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE212.HTM [19/1/2003 1:32:21]

Netlib

Next: Collected Algorithms of the Up: Software systems Previous: LEDA

Netlib

Netlib is an on-line repository of mathematical software that contains a large number of interesting codes, tables, and papers. Netlib is a compilation of resources from a variety of places, with fairly detailed indices and search mechanisms to help you find what is there. Netlib is important because of its breadth and ease of access. Whenever you need a specialized piece of mathematical software, you should look here first. There are three ways to access netlib: by e-mail, ftp, or WWW: ●

●

●

E-mail - Netlib provides an email server to send indices and sources on demand. To get an index, send e-mail to [email protected] with the words send index on its own line in the message. The index will provide a list of other files you can send for. The e-mail server and netlib in general are discussed in [DG87]. FTP - Connect by ftp to ftp.netlib.org. Log in as anonymous and use your e-mail address as password. Use ``ls'' to see the contents of a directory, ``cd'' to move to a different directory, and ``get'' to fetch the desired file. Type ``binary'' before ``get'' in order to ensure uncorrupted transmission, and ``quit'' to quit. Obtaining an index first can make it easier to move around. WWW - With your favorite browser, open the URL address http://www.netlib.org/ and prowl around to your heart's content. There is a forms index that permits searching based on keywords.

GAMS, the Guide to Available Mathematical Software, is an indexing service for Netlib and other related software repositories that can help you find what you want. Check it out at http://gams.nist.gov. GAMS is a service of the National Institute of Standards and Technology (NIST). Netlib pointers

●

Collected Algorithms of the ACM

file:///E|/BOOK/BOOK5/NODE213.HTM (1 of 2) [19/1/2003 1:32:21]

Netlib

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE213.HTM (2 of 2) [19/1/2003 1:32:21]

Collected Algorithms of the ACM

Next: The Stanford GraphBase Up: Netlib Previous: Netlib

Collected Algorithms of the ACM An early mechanism for the distribution of useful algorithm implementations was CALGO, the Collected Algorithms of the ACM. It first appeared in Communications of the ACM in 1960, covering such famous algorithms as Floyd's linear-time build heap algorithm. More recently, it has been the province of the ACM Transactions on Mathematical Software. Each algorithm/implementation is described in a brief journal article, with the implementation validated and collected. These implementations are maintained at http://www.acm.org/calgo/ and at Netlib. Over 750 algorithms have appeared to date. Most of the codes are in Fortran and are relevant to numerical computing, although several interesting combinatorial algorithms have slithered their way into CALGO. Since the implementations have been refereed, they are presumably more reliable than most readily available software.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE214.HTM [19/1/2003 1:32:22]

The Stanford GraphBase

Next: Combinatorica Up: Software systems Previous: Collected Algorithms of the

The Stanford GraphBase

The Stanford GraphBase is an interesting program for several reasons. First, it was composed as a ``literate program'', meaning that it was written to be read. If anybody's programs deserve to be read, it is Knuth's, and [Knu94] contains the full source code of the system. The programming language/environment is CWEB, which permits the mixing of text and code in particularly expressive ways. The GraphBase contains implementations of several important combinatorial algorithms, including matching, minimum spanning trees, and Voronoi diagrams, as well as specialized topics like constructing expander graphs and generating combinatorial objects. Finally, it contains programs for several recreational problems, including constructing word ladders (flour-floor-flood-blood-brood-broad-bread) and establishing dominance relations among football teams. Although the GraphBase is more fun to play with than LEDA, it is not really suited for building general applications on top of. The GraphBase is perhaps most useful as an instance generator for constructing a wide variety of graphs to serve as test data. It incorporates graphs derived from interactions of characters in famous novels, Roget's thesaurus, the Mona Lisa, and the economy of the United States. Further, because of its machine-independent random number generators, the GraphBase provides a way to construct random graphs that can be reconstructed elsewhere, making them perfect for experimental comparisons of algorithms. The Stanford GraphBase can be obtained by anonymous ftp from labrea.stanford.edu in the directory pub/sgb. It may be used freely, but the files may not be modified. Installing the GraphBase requires CWEB, which can be obtained by anonymous ftp from labrea.stanford.edu in the directory pub/cweb. GraphBase Implementation

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE215.HTM [19/1/2003 1:32:23]

Combinatorica

Next: Algorithm Animations with XTango Up: Software systems Previous: The Stanford GraphBase

Combinatorica

Combinatorica [Ski90] is a collection of over 230 algorithms for combinatorics and graph theory written in Mathematica. These routines have been designed to work together, enabling one to experiment with discrete structures and build prototype applications. Combinatorica has been widely used for both research and education. Although (in my totally unbiased opinion) Combinatorica is more comprehensive and better integrated than other libraries of combinatorial algorithms, it is also the slowest such system available. Credit for all of these properties is largely due to Mathematica, which provides a very high-level, functional, interpreted, and thus inefficient programming language. Combinatorica is best for finding quick solutions to small problems, and (if you can read Mathematica code) as a terse exposition of algorithms for translation into other languages. Combinatorica is included with the standard Mathematica distribution in the directory Packages/DiscreteMath/Combinatorica.m . It can also be obtained by anonymous ftp from ftp.cs.sunysb.edu in the pub/Combinatorica directory. Included on this site are certain extensions to Combinatorica and data sources such as the graphs of Graffiti. Combinatorica Implementation

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE216.HTM [19/1/2003 1:32:23]

Algorithm Animations with XTango

Next: Programs from Books Up: Software systems Previous: Combinatorica

Algorithm Animations with XTango

XTango [Sta90] is a general-purpose algorithm animation system, developed by John Stasko of Georgia Tech, that helps programmers develop color, real-time animations of their own algorithms and programs. Creating animations of your own implementations is useful both for pedagogical and debugging purposes. Included with the XTango distribution is a large collection of animations, several of which are quite interesting and enlightening to watch. The C language source for each animation is included. My favorites include animations of: ● ● ● ●

● ●

Data structures - including AVL/red-black trees, and binary/Fibonacci heaps. Sorting algorithms - including bubblesort, radixsort, quicksort, and shellsort. Backtracking - including graph coloring, and both the eight-queens and knight's tour problems. Geometric algorithms - including bin packing heuristics and the Graham scan/Jarvis march convex hull algorithms. Graph algorithms - including minimum spanning trees and shortest paths. String algorithms - including the Knuth-Morris-Pratt and Boyer-Moore algorithms.

Anybody studying these algorithms might well profit from playing with the animations. The source code for the animations contains C language implementations of these algorithms, although of uncertain quality. XTango is implemented on top of UNIX and X-windows. It is available via anonymous ftp from ftp.cc.gatech.edu in directory pub/people/stasko. Also available there is Polka, a C++ algorithm animation system that is particularly suited to animating parallel computations. POLKA provides its own high-level abstractions to make the creation of animations easier and faster than with many other systems. POLKA also includes an interactive front end called SAMBA that can be used to generate animations from any type of program that can generate ASCII. XTango Implementation

file:///E|/BOOK/BOOK5/NODE217.HTM (1 of 2) [19/1/2003 1:32:24]

Algorithm Animations with XTango

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE217.HTM (2 of 2) [19/1/2003 1:32:24]

Programs from Books

Next: Discrete Optimization Algorithms in Up: Software systems Previous: Algorithm Animations with XTango

Programs from Books Several books on algorithms include working implementations of the algorithms in a real programming language. Although these implementations are intended primarily for exposition, they can also be useful for computation. Since they are typically small and clean, they can prove the right foundation for simple applications. The most useful codes of this genre are described below. All are available from the algorithm repository, http://www.cs.sunysb.edu/ algorith.

● ● ● ● ● ●

Discrete Optimization Algorithms in Pascal Handbook of Data Structures and Algorithms Combinatorial Algorithms for Computers and Calculators Algorithms from P to NP Computational Geometry in C Algorithms in C++

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE218.HTM [19/1/2003 1:32:24]

Discrete Optimization Algorithms in Pascal

Next: Handbook of Data Structures Up: Programs from Books Previous: Programs from Books

Discrete Optimization Algorithms in Pascal

This is a collection of 28 programs for solving discrete optimization problems, appearing in the book by Syslo, Deo, and Kowalik [SDK83]. The package includes programs for integer and linear programming, the knapsack and set cover problems, traveling salesman, vertex coloring, and scheduling as well as standard network optimization problems. They have been made available from the algorithm repository WWW site, http://www.cs.sunysb.edu/ algorith. This package is noteworthy for the operations-research flavor of the problems and algorithms selected. The algorithms have been selected to solve problems, as opposed to for purely expository purposes. In [SDK83], a description of each algorithm and extensive set of references for each problem is provided, as well as execution times for each program on several instances on an early 1980s mainframe, an Amdahl 470 V/6. Implementations

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE219.HTM [19/1/2003 1:32:25]

Handbook of Data Structures and Algorithms

Next: Combinatorial Algorithms for Computers Up: Programs from Books Previous: Discrete Optimization Algorithms in

Handbook of Data Structures and Algorithms

The Handbook of Data Structures and Algorithms, by Gonnet and Baeza-Yates [GBY91], provides a comprehensive reference on fundamental data structures for searching and priority queues, and algorithms for sorting and text searching. The book covers this relatively small number of topics comprehensively, presenting most of the major and minor variations that appear in the literature. Perusing the book makes one aware of the tremendous range of data structures that have been developed and the intense level of analysis many of them have been subjected to. For each data structure or algorithm, a brief description is provided along with its asymptotic complexity and an extensive set of references. More distinctively, an implementation in C and/or Pascal is usually provided, along with experimental results comparing the performance of these implementations. The programs in [GBY91] are available at a very slick WWW site, http://www.dcc.uchile.cl/ rbaeza/handbook/hbook.html. Since many of the elementary data structures and sorting algorithms can be implemented concisely, most of the programs are very short. They are perhaps most useful as models, or as part of an experimental study to compare the performance on different data structures for a particular application. Handbook Implementations

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE220.HTM [19/1/2003 1:32:25]

Combinatorial Algorithms for Computers and Calculators

Next: Algorithms from P to Up: Programs from Books Previous: Handbook of Data Structures

Combinatorial Algorithms for Computers and Calculators

Nijenhuis and Wilf [NW78] specializes in algorithms for constructing basic combinatorial objects such as permutations, subsets, and partitions. Such algorithms are often very short, but they are hard to locate and usually surprisingly subtle. Fortran programs for all of the algorithms are provided, as well as a discussion of the theory behind each of them. The programs are usually short enough that it is reasonable to translate them directly into a more modern programming language, as I did in writing Combinatorica ). Both random and sequential generation algorithms are provided. Descriptions of more (see Section recent algorithms for several problems, without code, are provided in [Wil89]. These programs are now available from our algorithm repository WWW site. We tracked them down from Neil Sloane, who had them on a magnetic tape, while the original authors did not! In [NW78], Nijenhuis and Wilf set the proper standard of statistically testing the output distribution of each of the random generators to establish that they really appear uniform. We encourage you to do the same before using these programs, to verify that nothing has been lost in transit. Programs

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE221.HTM [19/1/2003 1:32:26]

Algorithms from P to NP

Next: Computational Geometry in C Up: Programs from Books Previous: Combinatorial Algorithms for Computers

Algorithms from P to NP

This algorithms text [MS91] distinguishes itself by including Pascal implementations of many algorithms, with careful experimental comparisons of different algorithms for such problems as sorting and minimum spanning tree, and heuristics for the traveling salesman problem. It provides a useful model for how to properly do empirical algorithm analysis. The programs themselves are probably best used as models. Interesting implementations include the eight-queens problem, plus fundamental graph and geometric algorithms. The programs in [MS91] have been made available by anonymous ftp from cs.unm.edu in directory /pub/moret_shapiro. Programs

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE222.HTM [19/1/2003 1:32:26]

Computational Geometry in C

Next: Algorithms in C++ Up: Programs from Books Previous: Algorithms from P to

Computational Geometry in C

O'Rourke [O'R94] is perhaps the best practical introduction to computational geometry available, because of its careful and correct C language implementations of the fundamental algorithms of computational geometry. Fundamental geometric primitives, convex hulls, triangulations, Voronoi diagrams, and motion planning are all included. Although they were implemented primarily for exposition rather than production use, they should be quite reliable. The codes are available from http://grendel.csc.smith.edu/ orourke/. Programs

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE223.HTM [19/1/2003 1:32:28]

Algorithms in C++

Next: Data Sources Up: Programs from Books Previous: Computational Geometry in C

Algorithms in C++

Sedgewick's popular algorithms text [Sed92] comes in several different language editions, including C, C++, and Modula-3. This book distinguishes itself through its use of algorithm animation and in its broad topic coverage, including numerical, string, and geometric algorithms. The language-specific parts of the text consist of many small code fragments, instead of full programs or subroutines. Thus they are best thought of as models, instead of working implementations. Still, the program fragments from the C++ edition have been made available from http://hegschool.aw.com/cseng/authors/sedgewick/algo-in-c++/algo-in-c++.html

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE224.HTM [19/1/2003 1:32:28]

Data Sources

Next: Textbooks Up: Algorithmic Resources Previous: Algorithms in C++

Data Sources It is often important to have interesting data to feed your algorithms, to serve as test data to ensure correctness or to compare different algorithms for raw speed. Finding good test data can be surprisingly difficult. Here are some pointers: ●

●

Combinatorica graphs - A modest collection of graphs is available from the Combinatorica ftp site. Particularly interesting are the 190 graphs of Graffiti [Faj87], a program that formulated conjectures in graph theory by studying the properties of these graphs. See Section for more information on Combinatorica. TSPLIB - This well-respected library of test instances for the traveling salesman problem is ) and by anonymous ftp from softlib.cs.rice.edu. TSPLIB available from Netlib (see Section instances are large, real-world graphs, derived from applications such as circuit boards and networks.

●

●

●

Stanford GraphBase - Discussed in Section , this suite of programs by Knuth provides portable generators for a wide variety of graphs. These include graphs arising from distance matrices, arts, and literature, as well as graphs of more theoretical interest. DIMACS Challenge data - A series of DIMACS Challenge workshops have focused on evaluating algorithm implementations of graph, logic, and data structure problems. Instance generators for each problem have been developed, with the focus on constructing difficult or representative test data. The products of the DIMACS Challenges are available from http://dimacs.rutgers.edu/. Algorithm Repository - The Algorithm Repository WWW site, http://www.cs.sunysb.edu/ algorith (and enclosed CD-ROM) contain data sources for a few of the implementation challenge exercises. In particular, we provide an airline routes data set, and a collection of names labeled by ethnicity.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE225.HTM [19/1/2003 1:32:29]

Textbooks

Next: On-Line Resources Up: Algorithmic Resources Previous: Data Sources

Textbooks There have emerged a number of excellent textbooks in the design and analysis of combinatorial algorithms. Below we point out several of our favorites. In this book, we have shied away from giving a detailed exposition or analysis of many algorithms, for our primary mission is to help the reader identify their problem and point them to solutions. The reader is encouraged to turn to these sources for more detail once they have found the name of what they are looking for. Only general algorithm books are discussed here. Books on specific subareas of algorithms are reviewed at the head of the relevant catalog chapter. ●

●

●

●

●

●

Cormen, Leiserson, and Rivest [CLR90] - This is the one (other) book on algorithms you must own, with its comprehensive treatment of most of the problems we discuss here, including data structures, graph algorithms, and seminumerical algorithms. Baase [Baa88] - This book is more accessible than [CLR90] for those without a strong mathematical background. It covers standard sorting, string, and graph algorithms, NPcompleteness, and more exotically, an introduction to parallel algorithms. Manber [Man89] - Built around the unconventional notion that induction is the fundamental paradigm of algorithm design, this book is especially good at teaching techniques for designing algorithms and has an outstanding collection of problems. Highly recommended. van Leeuwen [vL90b] - Not a textbook, but a collection of in-depth surveys on the state of the art in algorithms and computational complexity. Although the emphasis is on theoretical results, this book is perhaps the best single reference to point you to what is known about any given problem. Syslo, Deo, and Kowalik [SDK83] - This book includes printed Pascal implementations of 28 algorithms for discrete optimization problems, including mathematical programming, network optimization, and traditional operations research problems such as knapsack and TSP. Each algorithm is described in the book, and experimental timings (on a 1980s vintage machine) are . Despite its age, this provided. These codes are now available by ftp, as discussed in Section remains a useful reference, particularly with the programs now available on-line. Moret and Shapiro [MS91] - This algorithms text distinguishes itself by including Pascal implementations of all algorithms and by its careful experimental comparisons of different algorithms for such problems as sorting and minimum spanning tree. It provides a useful model for how to properly do empirical algorithm analysis. These programs are available by ftp - see Section for details.

file:///E|/BOOK/BOOK5/NODE226.HTM (1 of 2) [19/1/2003 1:32:30]

Textbooks ●

●

●

●

●

●

Knuth [Knu94] - This book presents the implementation of the Stanford GraphBase, a collection of programs for constructing different graphs and working with them. See Section for details. It is very intriguing to browse. Aho, Hopcroft, and Ullman [AHU74] - This was the first modern algorithms book, and it has had an enormous influence on how algorithms should be taught. Although it is now dated, it remains a useful guide to certain topics in vogue in the early 1970s, such as matrix multiplication, the fast Fourier transform, and arithmetic algorithms. A more elementary edition, focusing on data structures, is [AHU83]. Rawlins [Raw92] - This may well be the best self-study book available on algorithms. It is fun and inspiring, with built-in pauses so the reader can make sure they understand what is going on. The only drawback is a somewhat idiosyncratic set of topics, so you will miss certain important topics. But you can get that from here. Rawlins's book can teach you the proper mindset to think about algorithms. Papadimitriou and Steiglitz [PS82] - This book has more of an operations research emphasis than most algorithms texts, with a good coverage of mathematical programming, network flow, and combinatorial search. Lawler [Law76] - Particularly useful for its coverage of matroid theory, this book also provides a thorough treatment of the network flow, matching, and shortest path algorithms known by the mid1970s. Binstock and Rex [BR95] - Although not a textbook, it includes C language implementations of an idiosyncratic variety of algorithms for programmers. Disks containing the code are available for a modest fee. The most interesting implementations are string and pattern matching algorithms, time and date routines, an arbitrary-precision calculator, and a nice section on checksums and cyclicredundancy checks.

Next: On-Line Resources Up: Algorithmic Resources Previous: Data Sources Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE226.HTM (2 of 2) [19/1/2003 1:32:30]

On-Line Resources

Next: Literature Up: Algorithmic Resources Previous: Textbooks

On-Line Resources The Internet has proven to be a fantastic resource for people interested in algorithms, as it has for many other subjects. What follows is a highly selective list of the resources that I use most often, partitioned into references to literature, people, and software. All of these should be in the tool chest of every algorist.

● ● ●

Literature People Software

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE227.HTM [19/1/2003 1:32:30]

Literature

Next: People Up: On-Line Resources Previous: On-Line Resources

Literature There are many bibliographic sources available on the WWW, but the following I find indispensable. ●

●

●

●

Computer Science Bibliographies - This is a collection of over 700,000 references to papers and technical reports in Computer Science, beneath a sophisticated search engine. While there is much duplication, this is my first stop whenever I need to look something up. The primary site is http://liinwww.ira.uka.de/bibliography/index.html, although several mirror sites are in operation around the world. All references are provided in bibtex format. Joel Seiferas' paper.lst - References to over 55,000 papers and technical reports (at last count), mostly on algorithms and related areas of theoretical computer science. Each paper is reduced to a one-line format, which I find easy to grep through. It is available by anonymous ftp from ftp.cs.rochester.edu in /pub/u/joel, and a copy is included on the enclosed CD-ROM. Strongly recommended. Geom.bib - The complete bibliography on anything related to computational geometry, it references over 8,000 books, papers, and reports and includes detailed abstracts for many of them. Grep-ing through geom.bib is an amazingly efficient way to find out about previous work without leaving your office. It is available via anonymous ftp from ftp.cs.usask.ca, in file pub/geometry/geombib.tar.Z, and a copy is included on the enclosed CD-ROM. Usenet FAQ files - The straightest, most accurate dope about any topic is likely to be had in a USENET frequently asked questions (FAQ) file. Dedicated volunteers maintain FAQ files for many USENET newsgroups, many of which are quite detailed, and which always emphasize other on-line resources. Excellent FAQ files on algorithm-related topics such as cryptography, linear programming, and data compression are currently available. A collection of current FAQ files is maintained at ftp://rtfm.mit.edu/pub/usenet/news.answers.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE228.HTM [19/1/2003 1:32:31]

People

Next: Software Up: On-Line Resources Previous: Literature

People The easiest way to find the answer to any question is to ask someone who knows. Below, I describe some useful resources of contact information for experts in algorithms. E-mail and the WWW make it very easy to track down experts. In fact, maybe too easy. Please use this information responsibly, for many of them are my friends! Limit your queries to short, specific questions; and contact them only after you have checked the standard textbooks and references described in this section. ●

●

●

comp.theory - The USENET newsgroup comp.theory is the proper forum for on-line discussions on algorithms and related topics. Other appropriate newsgroups include comp.graphics.algorithms and sci.math. TCS Virtual Address Book - This WWW-page address book is particularly useful to track down the whereabouts of researchers in the design and analysis of algorithms. It is maintained by Ian Parbery and is available from the ACM SIGACT WWW page http://sigact.acm.org/sigact/. An excellent way to learn of recent progress on a particular problem is to ask a researcher who has previously published on the problem, or at least check out their WWW page. Algorithm Courses - A comprehensive page of pointers to algorithm courses on the WWW is maintained at http://www.cs.pitt.edu/ kirk/algorithmcourses/index.html. Lecture notes, problems, and solutions are typically available, with some of the material being amazingly thorough. Software associated with course projects is available from some of the sites.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE229.HTM [19/1/2003 1:32:31]

Software

Next: Professional Consulting Services Up: On-Line Resources Previous: People

Software There are many high-quality implementations of algorithms available on the WWW, if you know where to look. We have placed as many of them as we could on the CD-ROM enclosed with this book. My favorite on-line resources are: ●

●

●

●

The Stony Brook Algorithm Repository - This is the source of the software on the enclosed CDROM, organized according to the problem structure of this book. Local copies of most implementations are maintained at http://www.cs.sunysb.edu/ algorith, along with pointers to the latest version available from the original distribution site. Netlib - This is the standard resource on mathematical software, primarily numerical computation. for details. An enormous amount of material is available. See Section Directory of Computational Geometry Software - Maintained by Nina Amenta, this is the source for anything geometric. See http://www.geom.umn.edu/software/cglist/. Combinatorial Optimization codes - A very nice list of implementations of combinatorial algorithms is maintained by Jiefeng Xu at http://ucsu.colorado.edu/ xu/software.html. With an emphasis on operations research, there is a lot of useful stuff here.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE230.HTM [19/1/2003 1:32:32]

Professional Consulting Services

Next: References Up: Algorithmic Resources Previous: Software

Professional Consulting Services Algorist Technologies is a small consulting firm that provides its clients with short-term, expert help in algorithm design and implementation. Typically, an Algorist consultant is called in for 1 to 3 days worth of intensive, on-site discussion and analysis with the client's own development staff. Algorist has built an impressive record of performance improvements with several companies and applications. They provide longer-term consulting and contracting services as well. Call 212-580-9009 or email [email protected] for more information on services provided by Algorist Technologies. Algorist Technologies 312 West 92nd St. Suite 1A New York, NY 10025 http://www.algorist.com

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK5/NODE231.HTM [19/1/2003 1:32:32]

Index

Next: About this document Up: No Title Previous: Lecture 23 - approximation

Index This index provides fast access to important keywords and topics in the on-line Lecture Notes The menu below partitions the index entries by the first letter of the alphabet, for ease of access. A full document index is also provided. Be aware that the index pointer typically resides at the end of the relevant paragraph in the document, so we recommend scrolling once towards the front of the document before reading. A similar index has been provided for the book, which may also be of interest.

A B I

C D

E F

K L M N O

R S

T U

G H P

Q

V W

Complete Index (note: the complete index is large; it will take a bit of time to load)

Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR16/NODE24.HTM [19/1/2003 1:32:33]

Index A

Up: Index - All

Index: A above-below test , abracadabra abstract data types abstract graph type academic institutions - licensing acceptance-rejection method Ackerman function acyclic subgraph Ada adaptive compression algorithms Adaptive Simulated Annealing (ASA) addition address book, TCS adjacency list , adjacency matrix , adjacent swaps advice - caveat aesthetically pleasing drawings aggregate range queries agrep Aho-Corasick algorithm airline distance metric airline scheduling , algorist Algorist Technologies algorithm animation algorithm design algorithmic resources aligning DNA sequences alignment costs all-pairs shortest path ,

,

file:///E|/BOOK/BOOK6/INDEX_A.HTM (1 of 3) [19/1/2003 1:32:35]

Index A

alpha-beta pruning , alpha-shapes amortized analysis AMPL analog channel ancestor angle bisector animation - motion planning animation - sorting animation - systems annotated bibliography approximate nearest neighbor search , approximate string matching ,

,

,

approximate string matching - related problems , approximate substring matching approximation algorithms , approximation scheme , arbitrage Arbitrary-Precision Arithmetic arbitrary-precision arithmetic - geometry arbitrary-precision arithmetic - related problems architectural models area computations - applications area computations - triangles area minimization arm, robot around the world game Arrange , arrangement , , arrangement of objects arrangements of lines array array searching art gallery problems articulation vertex , file:///E|/BOOK/BOOK6/INDEX_A.HTM (2 of 3) [19/1/2003 1:32:35]

Index A

artists steal ASA ASCII aspect ratio assembly language , assignment problem associative operation asymmetric longest path problem asymmetric TSPs , asymptotic analysis atom smashing attitude of the algorithm designer attribute attribute - graph augmenting path augmenting path methods authentication protocol automorphisms average average-case analysis average-case complexity AVL tree Avogadro's number awk Axiom axis-oriented rectangles , axis-parallel planes

Algorithms Tue Jun 3 11:59:38 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_A.HTM (3 of 3) [19/1/2003 1:32:35]

Index B

Up: Index - All

Index: B B-tree , , , backpacker backsubstitution backtracking , , , backtracking - animations

,

,

,

,

backtracking - applications , backtracking - bandwidth problem balanced search tree ,

,

banded systems , bandersnatch problem bandwidth , bandwidth - matrix Bandwidth Reduction bandwidth reduction - backtracking bandwidth reduction - related problems bar codes base - arithmetic base - conversion base of logarithm Bellman-Ford algorithm , Berge's theorem best-case complexity Bible - searching the bibliographic databases biconnected components biconnected graphs , big Oh notation bijection binary heap

,

file:///E|/BOOK/BOOK6/INDEX_B.HTM (1 of 3) [19/1/2003 1:32:36]

,

,

Index B

binary representation - subsets binary search ,

,

binary search - applications , binary search - one-sided , binary search tree , , , binary search tree - applications binary search tree - computational experience Bin Packing bin packing - applications , bin packing - knapsack problem bin packing - related problems , biocomputing biology bipartite graph bipartite graph recognition bipartite incidence structures bipartite matching ,

,

,

bipartite matching - applications , bit-mapped images bit representation of graphs bit vector ,

,

,

bit vector - applications , blind man's algorithm block - set partition blossoms board evaluation function bookshelves Boolean logic minimization , Boolean matrix multiplication borrowing Boruvka's algorithm boss's delight boundaries bounded height priority queue

file:///E|/BOOK/BOOK6/INDEX_B.HTM (2 of 3) [19/1/2003 1:32:36]

Index B

bounding boxes Boyer-Moore algorithm brainstorming branch-and-bound search ,

,

breadth-first search , , , breadth-first search - applications bridge bridges of Königsberg Brook's theorem Brooks, Mel brush fire brute-force search bubblesort , bucketing techniques , , bucketing techniques - graphics bucket sort budget, fixed built-in random number generator buying fixed lots

Algorithms Tue Jun 3 11:59:39 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_B.HTM (3 of 3) [19/1/2003 1:32:36]

Index C

Up: Index - All

Index: C C++ ,

,

,

,

,

,

,

,

,

,

,

,

, , , , , , C++ templates cache Caesar shifts calculator, arithmetic Calendrical Calculations

,

,

,

,

,

,

,

call graph , canonically-labeled graphs canonical order , , CAP Carmichael numbers cars and tanks cartoons casino analysis casino poker catalog WWW site Catch-22 situation caveat CD-ROM , cdd

,

,

center vertex , , CGAL chain of matrices characters checksum chessboard coverage chess program , Chinese calendar Chinese postman problem file:///E|/BOOK/BOOK6/INDEX_C.HTM (1 of 6) [19/1/2003 1:32:40]

,

,

,

,

,

,

,

Index C

Chinese remainder theorem Christofides heuristic chromatic index chromatic number chromatic polynomials cipher circle circuit analysis circuit board assembly circuit board placement - simulated annealing circuit layout circuit schematic diagrams circuit testing circular embeddings C language ,

,

,

,

,

, , , , , , , classification classification - nearest-neighbor classifiers - neural networks clauses clipping Clique

,

, ,

, ,

clique - applications , clique - definition clique - hardness proof clique - related problems clock closest pair heuristic closest pair problem , closest point closure clothing - manufacturing cloudy days cluster clustered access cluster identification ,

,

file:///E|/BOOK/BOOK6/INDEX_C.HTM (2 of 6) [19/1/2003 1:32:40]

, ,

, ,

, ,

, ,

, ,

, ,

, ,

, ,

, ,

, ,

Index C

clustering , , co-NP coding theory coefficients cofactor method coin flip collapsing dense subgraphs Collected Algorithms of the ACM ,

,

,

,

,

,

,

,

,

,

,

,

,

,

, , , , , , , collection coloring graphs color interchange combinatorial generation algorithms combinatorial geometry combinatorial problems , Combinatorica ,

,

,

, , , , , , , Commentz-Walter algorithm commercial implementations committee committee - congressional Common Lisp common substrings communication in circuits

,

, ,

,

,

,

communications networks , comp.graphics.algorithms comp.theory compaction comparisons - minimizing compiler compiler construction compiler optimization , compiler optimization - performance complement complement graph completion time - minimum file:///E|/BOOK/BOOK6/INDEX_C.HTM (3 of 6) [19/1/2003 1:32:40]

,

,

,

,

,

,

,

,

,

,

Index C

complexity classes composite integer compress , compression compression - image computational biology computational complexity computational geometry computational number theory , computer algebra system , computer chess computer graphics computer graphics - applications , computer graphics - rendering computer vision concatenation - string concavities concavity elimination configurations configuration space conjugate gradient methods conjunctive normal form (CNF) connected components ,

,

,

connected components - related problems , connected graph connectivity , , consensus sequences consistent schedule Constrained and Unconstrained Optimization constrained and unconstrained optimization - related problems constrained Delaunay triangulation constrained optimization constrained optimization - related problems constraint elimination constraint satisfaction

file:///E|/BOOK/BOOK6/INDEX_C.HTM (4 of 6) [19/1/2003 1:32:40]

Index C

consulting services , container , context-free grammars Contig Assembly Program control systems - minimization convex decomposition , convex hull , Convex Hull convex hull - related problems , convex polygons convex polygons - intersection convex region convolution - polygon convolution - sequences cookbook cooling schedules coordinate transformations coplanar points copying a graph corporate ladder correctness - algorithm correlation function counterexample construction counting edges and vertices counting Eulerian cycles counting integer partitions counting linear extensions counting matchings counting paths , counting spanning trees courses, lecture notes covering polygons with convex pieces covering set elements CPLEX Cramer's rule CRC

file:///E|/BOOK/BOOK6/INDEX_C.HTM (5 of 6) [19/1/2003 1:32:40]

Index C

critical path method crossing number crossings Cryptography cryptography - keys cryptography - related problems , CS CSA cubic regions currency speculation curve fitting Cuthill-McKee algorithm

,

cut set , cutting plane methods , cutting stock problem CWEB cycle - shortest cycle breaking cycle detection , cycle length cycle notation cycle structure of permutations cyclic-redundancy check (CRC)

Algorithms Tue Jun 3 11:59:39 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_C.HTM (6 of 6) [19/1/2003 1:32:40]

Index D

Up: Index - All

Index: D DAG DAG - longest path in DAG - shortest path in data abstraction database algorithms database application database query optimization data compression Data Encryption Standard data filtering data records data structures , data structures - animations data transmission data validation Davenport-Schintzl sequences ,

,

Davis-Putnam procedure , day of the week calculation deadlock de Bruijn sequence , debugging graph algorithms debugging parallel programs debugging randomized algorithms debugging time debugging tools decimal arithmetic decompose space decomposing polygons deconvolution decrease-key decryption file:///E|/BOOK/BOOK6/INDEX_D.HTM (1 of 5) [19/1/2003 1:32:43]

Index D

Deep Blue defenestrate degeneracy degeneracy testing degenerate configuration degenerate system of equations degree, vertex , degree sequence degrees of freedom Delaunay triangulation , , Delaunay triangulation - applications deletion from binary search tree deletions - text deliveries and pickups delivery routing Democrat/Republican identification De Morgan's laws dense graphs , , densest sphere packing dense subgraph depth-first search ,

,

,

depth-first search - applications , depth-first search - backtracking derangement derivatives - automata derivatives - calculus DES descendent design process design rule checking determinant determinant - related problems Determinants and Permanents deterministic finite automata DFA diameter of a graph diameter of a point set

,

, ,

, ,

file:///E|/BOOK/BOOK6/INDEX_D.HTM (2 of 5) [19/1/2003 1:32:43]

, ,

, ,

Index D

Dictionaries dictionaries - related problems , dictionary , , dictionary - applications dictionary - related problems dictionary - searching DIEHARD diff - how it works digital geometry digital signatures Dijkstra's algorithm ,

,

DIMACS , , DIMACS Challenge data DIMACS Implementation Challenge , Dinic's algorithm directed acyclic graph , , directed cycle directed graph directed graphs - automata directory file structures disclaimer

,

,

discrete event simulation , Discrete Fourier Transform , discrete mathematics software discussion section disjoint paths disjoint set union disjoint subsets disjunctive networks disjunctive normal form , disk access disk drives , dispatching emergency vehicles , dispersion problems

file:///E|/BOOK/BOOK6/INDEX_D.HTM (3 of 5) [19/1/2003 1:32:43]

,

,

,

Index D

distance graph distance metrics distinguishable elements distributed computation distribution sort , divide and conquer ,

,

,

,

division , DNA DNA sequence comparisons DNA sequencing ,

,

dominance orderings , DOS file names double-precision arithmetic , , Douglas-Plucker algorithm drawing graphs - related problems Drawing Graphs Nicely drawing puzzles Drawing Trees drawing trees - related problems , driving time minimization drug discovery DSATUR dual graph , duality , duality transformations duplicate elimination duplicate elimination - graphs duplicate elimination - permutations duplicate keys dynamic convex hulls dynamic data structures , dynamic graph algorithms dynamic Huffman codes dynamic programming ,

,

,

,

,

file:///E|/BOOK/BOOK6/INDEX_D.HTM (4 of 5) [19/1/2003 1:32:43]

,

,

Index D

dynamic programming - applications , , dynamic programming - initialization dynamic programming - shortest paths dynamic programming - space efficiency

Algorithms Tue Jun 3 11:59:39 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_D.HTM (5 of 5) [19/1/2003 1:32:43]

Index E

Up: Index - All

Index: E eavesdropper eccentricity of a graph economics - applications to edge/vertex connectivity - related problems , Edge and Vertex Connectivity edge chromatic number edge coloring , edge coloring - applications edge coloring - related problems , edge cover , , edge disjoint paths edge flipping operation edge labeled graphs edge length edge tour edit distance , Edmond's algorithm efficiency of algorithms eight-queens problem electrical engineers electronic circuit analysis electronic circuits Electronic Frontier Foundation element uniqueness problem , elimination ordering ellipsoid algorithm elliptic-curve method embeddings - planar Emde Boas priority queue empirical results ,

,

file:///E|/BOOK/BOOK6/INDEX_E.HTM (1 of 3) [19/1/2003 1:32:45]

,

Index E

empirical results - heuristics empirical results - how to do empirical results - string matching employees to jobs - matching empty circle - largest empty rectangle enclosing boxes enclosing disk enclosing rectangle encryption energy function energy minimization , English language , English to French enumeration of spanning trees epsilon-moves equilateral triangle equivalence classes equivalence classes - automata states Erd•s-Gallai conditions error estimating closure sizes ethnic groups in Congress Euclid's algorithm Euclidean minimum spanning tree Euclidean traveling salesman Euler's formula Eulerian cycle - applications Eulerian cycle - line graphs Eulerian cycle - related problems , Eulerian Cycle / Chinese Postman Eulerian path evaluation function even-degree vertices even-length cycles event queue evolutionary tree

file:///E|/BOOK/BOOK6/INDEX_E.HTM (2 of 3) [19/1/2003 1:32:45]

Index E

exact cover problem exact string matching exam scheduling exercises ,

,

,

,

,

exhaustive search , exhaustive search - application exhaustive search - empirical results exhaustive search - subsets expanded obstacles approach expander graphs expected-time, linear experimental analysis - set cover experimental graph theory exponential-time algorithms , exponential distribution exponentiation , export restrictions external-memory sorting , external memory

Algorithms Tue Jun 3 11:59:40 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_E.HTM (3 of 3) [19/1/2003 1:32:45]

Index F

Up: Index - All

Index: F facets facility location , , Factoring and Primality Testing factoring and primality testing - related problems factoring integers - related problems factory location family tree , fan out minimization for networks FAQ file , , , farthest point Voronoi diagrams Fary's theorem faster computers fast Fourier transform fat cells fattening polygons feature sets Federal Sentencing Guidelines feedback edge set Feedback Edge/Vertex Set feedback edge/vertex set - related problems Fermat Fermat's theorem Ferrer's diagram FFT , FFTPACK fgrep Fibonacci heap , , , Fibonacci numbers FIFO file difference comparison file directory trees file:///E|/BOOK/BOOK6/INDEX_F.HTM (1 of 3) [19/1/2003 1:32:46]

Index F

file layout filtering outlying elements filtering signals final examination financial constraints find operation finite automata finite automata minimization finite element analysis Finite State Machine Minimization FIRE Engine firehouse first-fit - decreasing first in, first out fixed degree sequence graphs FLAP flat-earth model Fleury's algorithm flight crew scheduling floating-point arithmetic Floyd's algorithm , football program football scheduling

,

,

Fortran ,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

, , , , Fortune's algorithm

four-color problem , Fourier transform - applications Fourier transform - multiplication Fourier transform - related problems four Russians algorithm , fragment ordering fraud - tax freedom to hang yourself free space free trees frequency distribution

,

file:///E|/BOOK/BOOK6/INDEX_F.HTM (2 of 3) [19/1/2003 1:32:46]

,

,

,

,

,

,

,

,

Index F

frequency domain friend-or-foe identification friendship graph , ftp - instructions function interpolation furniture moving furthest-point insertion heuristic furthest-site diagrams furthest-site Voronoi vertices future events

Algorithms Tue Jun 3 11:59:40 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_F.HTM (3 of 3) [19/1/2003 1:32:46]

Index G

Up: Index - All

Index: G game-tree search game-tree search - parallel games directory GAMS , gaps between primes garbage trucks Garey and Johnson Gates, William Gaussian distribution , Gaussian elimination , Genbank searching Generating Graphs Generating Partitions generating partitions - related problems , Generating Permutations

,

,

generating permutations - related problems , Generating Subsets generating subsets - applications generating subsets - related problems ,

,

genetic algorithms , , , Genocop geographic information systems Geolab geom.bib , geometric algorithms - animations geometric data structure geometric degeneracy geometric graphs geometric primitives - related problems

file:///E|/BOOK/BOOK6/INDEX_G.HTM (1 of 3) [19/1/2003 1:32:48]

,

,

,

,

Index G

geometric shortest path , geometric spanning tree geometric Steiner tree geometric traveling salesman problem geometric TSP GEOMPACK , Gettysburg Address Gibbs-Poole-Stockmeyer algorithm gift-wrapping algorithm Gilbert and Pollak conjecture Gingrich, Newt girth , glimpse global optimization Graffiti Graffiti - graphs of , Graham scan Grail graph algorithms , graph algorithms - animations graph algorithms - bandwidth problem GraphBase , , graph complement

,

,

,

,

,

graph data structures , graph data structures - applications graph data structures - LEDA , graph density graph drawings - clutter GraphEd , , , graph embedding graphical enumeration graphic partitions Graphics Gems graphics plotter graph isomorphism ,

,

file:///E|/BOOK/BOOK6/INDEX_G.HTM (2 of 3) [19/1/2003 1:32:48]

,

,

,

,

,

,

Index G

graph isomorphism - related problems , graph partition , Graph Partition graph partition - related problems , graph products graphs graph theory graph theory packages graph traversal GraphViz

,

Gray code , greatest common divisor greedy heuristic , , , , , , greedy heuristic - Huffman codes greedy heuristic - minimum spanning trees greedy heuristic - superstrings Gregorian calendar grid embeddings grid file , group - automorphism growth rates guarantees - importance of guarding art galleries Guide to Available Mathematical Software gzip ,

Algorithms Tue Jun 3 11:59:40 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_G.HTM (3 of 3) [19/1/2003 1:32:48]

Index H

Up: Index - All

Index: H had-sex-with graph , half-space intersection Hamiltionian cycle - hypercube Hamiltonian cycle , , , Hamiltonian Cycle Hamiltonian cycle - applications Hamiltonian cycle - counting Hamiltonian cycle - hardness proof Hamiltonian cycle - line graphs Hamiltonian cycle - related problems , Hamiltonian path Hamiltonian path - applications Hamming distance hardness of approximation hardware arithmetic hardware design applications hardware implementation hash function hash tables hash tables - computational experience hash tables - size Hausdorff distance heap heapsort , , , heard-of graph heart-lung machine heating ducts Hebrew calendar Hertel-Mehlhorn heuristic heuristics , , heuristics - empirical results file:///E|/BOOK/BOOK6/INDEX_H.HTM (1 of 3) [19/1/2003 1:32:50]

Index H

hidden-surface elimination , hierarchical decomposition , hierarchical drawings hierarchical graph structures , hierarchy high-precision arithmetic - need for high-precision arithmetic - related problems , higher-dimensional data structures higher-dimensional geometry , high school algebra high school cliques hill climbing historical objects

,

history , history - cryptography history - graph theory hitting set HIV virus homeomorphism homophones horizon Horner's rule , How to Solve It ht://Dig hub site Huffman codes Hull Human Genome Initiative Hungarian algorithm hypercube , hypergraph , , hyperlinks, WWW hyperplanes hypertext layout

file:///E|/BOOK/BOOK6/INDEX_H.HTM (2 of 3) [19/1/2003 1:32:50]

Index H

Algorithms Tue Jun 3 11:59:41 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_H.HTM (3 of 3) [19/1/2003 1:32:50]

Index I

Up: Index - All

Index: I I_COLLIDE identical graphs IEEE Data Compression Conference image compression , image data image features image filtering image processing image segmentation image simplification

,

,

implementation challenge, DIMACS , implementation challenges , , , , , implementation complexity implementations, caveats implementation wanted implicit binary tree impress your friends algorithms in-circle test incidence matrices inconsistent linear equations increasing subsequences incremental algorithms incremental change methods incremental insertion algorithm incremental insertion algorithms - arrangements incremental insertion algorithms - coloring incremental insertion algorithms - graph drawing incremental insertion algorithms - sorting incremental insertion algorithms - suffix trees incremental insertion algorithms - TSP independent set , file:///E|/BOOK/BOOK6/INDEX_I.HTM (1 of 3) [19/1/2003 1:32:51]

Index I

independent set - alternate formulations independent set - hardness proof independent set - related problems , independent set - simulated annealing index - how to use

,

,

induced subgraph , induced subgraph isomorphism induction for algorithm design inequivalence of programs with assignments information retrieval information theory input/output graphics insertion into binary search tree insertions - text insertion sort , , , inside/outside polygon instance - definition instance generator integer arithmetic integer factorization , integer partition , , integer programming

,

integer programming - applications , Integer programming - hardness proof integer programming - related problems integrality constraints interfering tasks interior-point methods Internal Revenue Service (IRS) Internet , , interpolation search intersection - halfspaces intersection - set Intersection Detection intersection detection - applications

file:///E|/BOOK/BOOK6/INDEX_I.HTM (2 of 3) [19/1/2003 1:32:51]

Index I

intersection detection - related problems , intersection point interview scheduling invariant - graph inverse Ackerman function inverse Fourier transform inverse matrix , inverse operations inversions Islamic calendar isolated vertex isomorphism isomorphism - graph isomorphism-complete iterative methods - linear systems

Algorithms Tue Jun 3 11:59:41 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_I.HTM (3 of 3) [19/1/2003 1:32:51]

Index J

Up: Index - All

Index: J jigsaw puzzle job-shop scheduling job matching Job Scheduling Journal of Algorithms JPEG , Julian calendar

Algorithms Tue Jun 3 11:59:41 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_J.HTM [19/1/2003 1:32:52]

Index K

Up: Index - All

Index: K Königsberg k-subset - applications k-subsets Karatsuba's algorithm Karazanov's algorithm Karmarkar's algorithm Karp-Rabin algorithm Kd-Trees , kd-trees - applications kd-trees - related problems , Kepler conjecture

,

Kernighan-Lin heuristic , key length , key search Kirchhoff's laws knapsack knapsack problem Knapsack Problem knapsack problem - applications knapsack problem - related problems knight's tour problem Knuth-Morris-Pratt algorithm Kolmogorov complexity Kruskal's algorithm , , , kth-order Voronoi diagrams kth-shortest path Kuratowski's theorem

,

file:///E|/BOOK/BOOK6/INDEX_K.HTM (1 of 2) [19/1/2003 1:32:52]

Index K

Algorithms Tue Jun 3 11:59:41 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_K.HTM (2 of 2) [19/1/2003 1:32:52]

Index L

Up: Index - All

Index: L labeled graphs , labeling maps label placement labels language pattern matching LAPACK , large graphs - representation largest element last in, first out layered printed circuit boards Lazy adjacency matrix LCA - least common ancestor leap year least-squares curve fitting least common ancestor leaves - tree LEDA ,

,

,

, , , , left-right test left-to-right ordering

,

, ,

, ,

, ,

, ,

, ,

Lempel-Ziv algorithms , Lenstra's elliptic curve method lexicographic order , , , libraries licensing arrangements LIFO lifting-map construction line-point duality linear-time graph algorithms linear algebra , file:///E|/BOOK/BOOK6/INDEX_L.HTM (1 of 3) [19/1/2003 1:32:54]

, ,

,

,

,

,

,

,

,

,

,

Index L

linear congruential generator linear constraint satisfaction linear extension linear interpolation search linear partitioning linear programming Linear Programming linear programming - models linear programming - related problems , linear programming - relaxation linear programming - special cases line arrangements line graph , line intersection , line segment intersection line segment Voronoi diagram LINK , link distance , linked lists vs. arrays , LINPACK , , LISP list searching literate program locality of reference , local optima locations logarithms , logic minimization logic problems logic programming long division longest common prefix longest common substring , longest common substring - related problems ,

file:///E|/BOOK/BOOK6/INDEX_L.HTM (2 of 3) [19/1/2003 1:32:54]

Index L

longest cycle , longest increasing subsequence , longest path , longest path - DAG long keys loop , lossless encodings lossy encodings lottery problems Lotto problem low-degree spanning tree , low-dimensional linear programming lower bound , , , lower bound - range searching lower bound - sorting lower triangular matrix lp_solve LU-decomposition , lunar calendar LZW algorithm ,

Algorithms Tue Jun 3 11:59:41 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_L.HTM (3 of 3) [19/1/2003 1:32:54]

Index M

Up: Index - All

Index: M machine-independent random number generator machine clock Macsyma mafia magnetic tape mail routing maintaining arrangements - related problems , Maintaining Line Arrangements makeg manufacturing applications , map labeling Maple map making marriage problems MAT , matching , matching - applications matching - dual to matching - number of perfect matching - related problems , matching shapes

,

,

,

Mathematica ,

,

,

,

,

,

,

, , , , , mathematical notation

,

,

,

,

mathematical programming , mathematical software - netlib matrix-tree theorem matrix bandwidth matrix compression

file:///E|/BOOK/BOOK6/INDEX_M.HTM (1 of 4) [19/1/2003 1:32:57]

,

,

,

,

,

,

,

,

,

,

,

Index M

matrix inversion , matrix multiplication Matrix Multiplication matrix multiplication - applications matrix multiplication - related problems matroids max-cut max-flow, min-cut theorem maxima maximal clique maximal matching maximum-cardinality matchings maximum acyclic subgraph maximum cut - simulated annealing maze , McDonald's restaurants MD5 mean Mechanical computers mechanical truss analysis medial-axis transform , Medial-Axis Transformation median - application Median and Selection medical residents to hospitals - matching memory accesses mems Menger's theorem mergesort , , merging subsets merging tapes

,

mesh generation , Metaphone Metropolis algorithm middle square method millennium bug Miller-Rabin algorithm file:///E|/BOOK/BOOK6/INDEX_M.HTM (2 of 4) [19/1/2003 1:32:57]

Index M

mindset minima minimax search minimizing automata minimum-change order minimum change order - subsets minimum cut minimum equivalent digraph minimum spanning tree , Minimum Spanning Tree

,

,

,

minimum spanning tree - applications , minimum spanning tree - drawing minimum spanning tree - related problems , minimum weight triangulation Minkowski metric Minkowski sum , Minkowski sum - applications Minkowski sum - related problems , MIX assembly language mixed-integer programming mixed graphs mode , mode-switching modeling modeling algorithm problems modeling graph problems models of computation Modula-3 modular arithmetic molecular docking molecular sequence data Mona Lisa , monotone decomposition monotone polygons Monte Carlo techniques ,

file:///E|/BOOK/BOOK6/INDEX_M.HTM (3 of 4) [19/1/2003 1:32:57]

,

Index M

month and year morphing motion planning Motion Planning motion planning - related problems , motion planning - shape simplification mountain climbing

,

move to front rule , moving furniture MPEG , multicommodity flow multigraph multiple knapsacks multiple precision arithmetic multiple sequence alignment multiplication , multiplication, matrix multiplication algorithms multiset , musical scales

Algorithms Tue Jun 3 11:59:42 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_M.HTM (4 of 4) [19/1/2003 1:32:57]

Index N

Up: Index - All

Index: N name variations, recognizing naming concepts nanosecond national debt National Football League (NFL) National Security Agency (NSA) nauty , NC - Nick's class nearest-neighbor heuristic nearest neighbor - related problems nearest neighbor graph , nearest neighbor search ,

,

nearest neighbor search - related problems , negation negative-cost cycle negative cost edges NEOS , Netlib , , , , , , , , network Network-Enabled Optimization System

,

,

network design , , network design - minimum spanning tree network flow , Network Flow network flow - applications , network flow - related problems ,

,

,

network reliability , neural networks

file:///E|/BOOK/BOOK6/INDEX_N.HTM (1 of 3) [19/1/2003 1:32:59]

,

,

Index N

neural networks - classification neural networks - coloring , newsgroups next subset Nobel Prize , noisy channels noisy images , nonapproximability results noncrossing drawing nondeterministic automata nonEuclidean distance metrics nonnumerical problems nonorthogonal kd-tree nonself intersecting polygons nonuniform access normal distribution notorious NP-complete problem NP , NP-completeness NP-completeness - definition of NP-completeness - proving NP-completeness - theory of NP-complete problem , , , NP-complete problem - bandwidth NP-complete problem - crossing number NP-complete problem - NFA minimization NP-complete problem - satisfiability NP-complete problem - set packing NP-complete problem - superstrings NP-complete problem - tetrahedralization NP-complete problem - tree drawing NP-complete problem - trie minimization NP-hard problems nuclear fission number field sieve number theory ,

file:///E|/BOOK/BOOK6/INDEX_N.HTM (2 of 3) [19/1/2003 1:32:59]

Index N

numerical analysis numerical precision Numerical Recipes , numerical root finding numerical stability ,

Algorithms Tue Jun 3 11:59:42 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_N.HTM (3 of 3) [19/1/2003 1:32:59]

Index O

Up: Index - All

Index: O O-notation objective function obstacle-filled rooms OCR octtree odd-degree vertices odd-length cycles , off-line problem oligonucleotide arrays on-line algorithm resources on-line problem one-sided binary search , OpenGL graphics library operations research , optical character recognition , , , optical character recognition - system testing optimal binary search trees optimization problems ordered set ordering , order statistics organic graphs organ transplant orthogonal planes orthogonal polyline drawings orthogonal range query , outerplanar graphs outlying elements output-sensitive algorithms overdetermined linear systems

file:///E|/BOOK/BOOK6/INDEX_O.HTM (1 of 2) [19/1/2003 1:33:00]

Index O

overlap graph overpasses - highway Oxford English Dictionary

Algorithms Tue Jun 3 11:59:42 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_O.HTM (2 of 2) [19/1/2003 1:33:00]

Index P

Up: Index - All

Index: P P P-completeness p-tree packaging packaging applications packing vs. covering paging , pagoda pairing heap , palindrome paradigms of algorithms design parallel algorithms , parallel algorithms - graphs parallel algorithms - visualization parallel lines parallel processor scheduling paranoia level parenthesization PARI , parse trees parsing partial key search partial order , partitioning automata states partitioning point sets partitioning polygons into convex pieces partitioning problems partition problem party affiliations Pascal ,

,

,

,

,

,

,

,

file:///E|/BOOK/BOOK6/INDEX_P.HTM (1 of 5) [19/1/2003 1:33:02]

,

,

,

,

,

,

,

,

,

,

,

Index P

,

,

,

,

,

,

,

,

,

password , patented algorithms path path generation - backtracking path planning paths - counting , Patricia trie pattern matching ,

,

,

pattern recognition , pattern recognition - automata patterns Pat tree PDF-417 penalty functions perfect hashing perfect matching performance bottlenecks performance guarantee performance in practice period periodicities perl permanent permutation , permutation comparisons permutation generation permutation generation - backtracking perpendicular bisector personality conflicts - avoiding PERT/CPM Petersen graph PGP , , phone company PHYLIP

file:///E|/BOOK/BOOK6/INDEX_P.HTM (2 of 5) [19/1/2003 1:33:02]

,

Index P

phylogenic tree , piano mover's problem Picasso, P. , pieces of a graph pilots pink panther pivoting rules , pixel geometry , pLab planar drawings , planar drawings - related problems planar graph , planar graph - clique planar graph - coloring planar graph - instances planar graph - isomorphism Planarity Detection and Embedding planarity testing - related problems planar separators planar subdivisions planar sweep algorithms plumbing point-spread function point distributions pointer manipulation point in polygon point location , point location - related problems , , point robots points point set clusters Poisson distribution Polka polygonal data structure Polygon Partitioning polygon partitioning - related problems

,

file:///E|/BOOK/BOOK6/INDEX_P.HTM (3 of 5) [19/1/2003 1:33:02]

Index P

polygons polygon triangulation polyhedral simplification polyline graph drawings polynomial-time approximation scheme polynomial-time problems polynomial evaluation polynomial multiplication poor thin people popular keys porting code POSIT positions position tree potential function power diagrams power set powers of graphs Prüfer codes , precedence-constrainted scheduling precedence constraints , precision preemptive scheduling prefix - string preflow-push methods preprocessing - graph algorithms presortedness measures Pretty Good Privacy previous subset PRF price-per-pound Prim's algorithm , primality testing , prime number prime number theorem principle of optimality

file:///E|/BOOK/BOOK6/INDEX_P.HTM (4 of 5) [19/1/2003 1:33:02]

Index P

printed circuit boards , printing a graph priority queues , priority queues - applications , , priority queues - arithmetic model priority queues - related problems problem - definition problem-specific algorithms problem descriptions problem instance

,

,

problem solving techniques , procedure call overhead producer/consumer sectors profit maximization Program Evaluation and Review Technique program libraries programming languages programming time , program structure Prolog proof of correctness propagating consequences propositional logic protocol pruning - backtracking , pseudocode pseudorandom numbers psychic lotto prediction public key cryptography ,

,

,

Algorithms Tue Jun 3 11:59:42 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_P.HTM (5 of 5) [19/1/2003 1:33:02]

Index Q

Up: Index - All

Index: Q Qhull , , , quadratic-sieve method quadratic programming quadtree quality triangulations questions queue , queue - applications quicksort , , , quicksort - applications

Algorithms Tue Jun 3 11:59:42 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_Q.HTM [19/1/2003 1:33:03]

Index R

Up: Index - All

Index: R rabbits radial embeddings radio stations radius of a graph radix sort , RAM RAM model of computation Random Access Machine (RAM) random generation - testing random graph theory , random graphs - generation randomization randomized algorithms , , randomized data structures

,

,

randomized incremental algorithms , randomized quicksort randomized search - applications

,

,

random number generation , , random number generation - related problems random permutations , random perturbations random sampling - applications random search tree random subset Ranger ,

,

range search ,

,

range search - related problems , ranked embedding ranking and unranking operations ,

,

file:///E|/BOOK/BOOK6/INDEX_R.HTM (1 of 3) [19/1/2003 1:33:05]

Index R

ranking combinatorial objects ranking permutations ranking subsets RAPID rasterized images rational arithmetic ray shooting reachability problems rebalancing recommendations, caveat rectangle rectilinear Steiner tree recurrence relations recurrence relations - evaluation recursion - applications red-black tree reduction , reduction - direction of reflex vertices region of influence regions regions formed by lines register allocation regular expressions , relationship reliability, network repeated vertices representative selection Republican sex offenders resource allocation , resources - algorithm restricted growth function retrieval , reverse-search algorithms Right Stuff, The riots ensuing Rivest-Shamir-Adelman file:///E|/BOOK/BOOK6/INDEX_R.HTM (2 of 3) [19/1/2003 1:33:05]

Index R

road network ,

,

robot assembly , robot motion planning , , robust geometric computations Robust Geometric Primitives Roget's Thesaurus , rooted tree , root finding algorithms , rotating-calipers method rotation rotation - polygon roulette wheels round-off error round-off errors RSA-129

,

RSA algorithm , , rules of algorithm design run-length coding

Algorithms Tue Jun 3 11:59:42 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_R.HTM (3 of 3) [19/1/2003 1:33:05]

Index S

Up: Index - All

Index: S s-t connectivity safe cracker sequence satisfiability , satisfiability - related problems , satisfying constraints sato scaling , scanner, OCR scattered subsequences scene interpolation scheduling , scheduling - precedence constraints scheduling - related problems , scheduling problems

,

Scheme , schoolhouse method sci.math scientific computing , Searching

,

searching - related problems , search space search time minimization - magnetic media search tree , secondary key secondary storage devices secure hashing function security , seed

file:///E|/BOOK/BOOK6/INDEX_S.HTM (1 of 8) [19/1/2003 1:33:09]

Index S

segmentation , segment intersection selection , , selection - subsets selection sort , self-intersecting polygons self-organizing list , self-organizing tree , self-study textbook semi-exhaustive greedy algorithm semidefinite programming sentence structure separation problems separator theorems sequence sequencing by hybridization sequencing permutations sequential search , set set algorithms Set Cover , , set cover - applications set cover - exact set cover - related problems ,

,

,

Set Data Structures , set data structures - applications set data structures - related problems Set Packing , set packing - related problems , set partition , sex offenders, Republican shape of a point set shape representation shapes file:///E|/BOOK/BOOK6/INDEX_S.HTM (2 of 8) [19/1/2003 1:33:09]

Index S

Shape Similarity shape simplification shape simplification - applications , shellsort , Shifflett shift-register sequences shipping applications shipping problems Shortest Common Superstring shortest common superstring - related problems , shortest cycle shortest path , Shortest Path

,

,

shortest path - applications , shortest path - definition shortest path - geometric , shortest path - related problems , shortest path matrix shotgun sequencing shuffling sieving devices - mechanical SIGACT sign - determinant sign - permutation signal processing signal propagation minimization

,

,

Sim++ , SimPack , simple cycle simple graph simple polygon - construction simple polygons simplex method simplicial complex simplicity testing

file:///E|/BOOK/BOOK6/INDEX_S.HTM (3 of 8) [19/1/2003 1:33:09]

,

,

,

Index S

simplification envelopes Simplifying Polygons simplifying polygons - related problems simulated annealing , , , , simulated annealing - satisfiability simulated annealing - theory simulations simulations - accuracy sin, state of sine functions

,

,

single-precision numbers , single-source shortest path singular matrix , sinks - multiple sink vertex sites size of graph skeleton , skewed distribution Skiena, Len , skiing skinny triangles skip list slab method slack variables small edge weights smallest element smallest enclosing circle problem Smith Society smoothing , smoothness SNNS snow plows soap films software engineering software tools

file:///E|/BOOK/BOOK6/INDEX_S.HTM (4 of 8) [19/1/2003 1:33:09]

,

,

,

,

,

Index S

solar year Solving Linear Equations solving linear equations - related problems ,

,

sorted array , sorted linked list , sorting , , sorting - applications sorting - animations sorting - applications , sorting - cost of sorting - rationales for sorting - related problems , sorting - strings sound-alike strings

,

,

,

Soundex , sources - multiple source vertex space-efficient encodings space decomposition space minimization - digraphs space minimization - string matching spanning tree SPARE Parts sparse graph , , sparse matrices sparse matrices - compression sparse subset sparse systems sparsification spatial data structure special-purpose hardware speech recognition speedup - parallel spelling correction , sphere packing

,

file:///E|/BOOK/BOOK6/INDEX_S.HTM (5 of 8) [19/1/2003 1:33:09]

,

Index S

spikes Spinout puzzle spiral polygon splay tree , SPLIB splicing cycles splines split-and-merge algorithm spreadsheet updates spring embedding heuristics , square of a graph , square root of a graph square roots stable marriages , stable sorting stack , stack - applications stack size standard form Stanford GraphBase , , , star-shaped polygon decomposition state elimination, automata static tables statistical significance statistics steepest descent methods Steiner points Steiner ratio Steiner Tree Steiner tree - related problems Steiner vertices stock exchange stock picking Stony Brook Algorithm Repository Stony Brook class projects ,

file:///E|/BOOK/BOOK6/INDEX_S.HTM (6 of 8) [19/1/2003 1:33:09]

Index S

straight-line graph drawings , Strassen's algorithm , , , strategy strength of a graph string string algorithms string algorithms - animations string data structures ,

,

,

String Matching , string matching - related problems , string overlaps strings strings - combinatorial strings - generating strongly-connected graphs strongly connected components

,

,

strongly connected graphs , Stuttgart Neural Network Simulator subgraph isomorphism subgraph isomorphism - applications subroutine call overhead , subset subset generation subset generation - backtracking subset sum problem substitution cipher substitutions, text substring matching , subtraction suffix array suffix trees ,

,

suffix trees - applications , suffix trees - computational experience suffix trees - related problems ,

file:///E|/BOOK/BOOK6/INDEX_S.HTM (7 of 8) [19/1/2003 1:33:09]

Index S

Suffix Trees and Arrays sunny days supercomputer superstrings - shortest common surface interpolation surface structures swap elements swapping sweepline algorithms , , symbolic computation symbolic set representation Symbol Technologies symmetric difference symmetry detection symmetry removal

Algorithms Tue Jun 3 11:59:43 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_S.HTM (8 of 8) [19/1/2003 1:33:09]

Index T

Up: Index - All

Index: T tables tabu search tactics tail recursion take home lessons tape drive tax fraud taxonomy technical skills telephone books , telephone dialing

,

terrorist , test data testing planarity test pilots tetrahedralization text textbooks , text compression , Text Compression text compression - related problems ,

,

text data structures , text processing algorithms text searching with errors thermodynamics thinning thinning - applications thinning - related problems , three-points-on-a-line tight bound file:///E|/BOOK/BOOK6/INDEX_T.HTM (1 of 3) [19/1/2003 1:33:11]

Index T

time-series analysis time slot scheduling Toeplitz matrices tool path optimization topological sorting , topological sorting - applications , topological sorting - related problems , topological sweep tour traffic light scheduling transition matrix transitive closure Transitive Closure and Reduction transitive reduction translation - polygon transmitter power transportation problems , transposition trapezoidal decomposition traveling salesman ,

,

,

,

traveling salesman - applications , traveling salesman - approximation algorithms traveling salesman - decision problem traveling salesman - dynamic programming traveling salesman - related problems , traveling salesman - simulated annealing Traveling Salesman Problem treap tree identification

,

trees , trees - acyclic graphs trees - detection trees - drawings trees - generation trees - hard problem in trees - independent set file:///E|/BOOK/BOOK6/INDEX_T.HTM (2 of 3) [19/1/2003 1:33:11]

,

Index T

trees - matching trees - partition trial division Triangle triangle inequality , triangle refinement method triangle strips , triangulated surfaces Triangulation triangulation - applications , , triangulation - minimum weight

,

triangulation - related problems , triconnected components trie , trigram statistics TSP tsp_solve TSPLIB , turnpike reconstruction problem twenty questions two-coloring

Algorithms Tue Jun 3 11:59:43 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_T.HTM (3 of 3) [19/1/2003 1:33:11]

Index U

Up: Index - All

Index: U unbounded search , unconstrained optimization , , unconstrained optimization - related problems undirected graph uniform distribution , , union, set union-find data structure union-find data structure - applications union of polygons union of polygons - applications unit cube unit sphere universal set unknown data structures unlabeled graphs , unranking combinatorial objects unranking permutations unranking subsets unsorted array unsorted list unweighted graphs - spanning trees upper bound upper triangular matrix Utah

Algorithms Tue Jun 3 11:59:43 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_U.HTM [19/1/2003 1:33:12]

Index V

Up: Index - All

Index: V validation Vancouver Stock Exchange Vandermonde matrices vantage point tree variable elimination variable length encodings vector quantification vector sums Vertex Coloring , , , vertex coloring - applications vertex coloring - bipartite graphs vertex coloring - related problems ,

,

vertex cover , vertex cover - approximation algorithm vertex cover - hardness proof , vertex cover - related problems ,

,

vertex degree , vertex disjoint paths video - algorithm animation video compression , virtual memory , , virtual memory - algorithms virtual memory - performance virtual reality applications visibility graphs , Viterbi algorithm Vizing's theorem , VLSI circuit layout ,

file:///E|/BOOK/BOOK6/INDEX_V.HTM (1 of 2) [19/1/2003 1:33:13]

Index V

VLSI design problems volume computations , von Emde Boas queue von Neumann, J. Voronoi diagram Voronoi Diagrams Voronoi diagrams - nearest neighbor search Voronoi diagrams - related problems ,

,

Algorithms Tue Jun 3 11:59:43 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_V.HTM (2 of 2) [19/1/2003 1:33:13]

,

,

Index W

Up: Index - All

Index: W walk-through Waring's problem , Warshall's algorithm war story , water pipes wavelets

,

,

,

,

,

,

weakly-connected graphs , web weighted graphs, applications Winograd's algorithm wire length minimization wiring layout problems word ladders , worker assignment - scheduling world's record TSP worst-case complexity WWW site

Algorithms Tue Jun 3 11:59:43 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_W.HTM [19/1/2003 1:33:14]

,

,

Index X

Up: Index - All

Index: X X-windows Xerox machines - scheduling XRLF XTango ,

,

,

,

,

,

,

Algorithms Tue Jun 3 11:59:43 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_X.HTM [19/1/2003 1:33:15]

,

,

,

,

,

,

,

,

,

,

,

,

Index Y

Up: Index - All

Index: Y Young tableaux

Algorithms Tue Jun 3 11:59:44 EDT 1997

file:///E|/BOOK/BOOK6/INDEX_Y.HTM [19/1/2003 1:33:15]

Index Z

Up: Index - All

Index: Z zero-knowledge proofs Zipf's law zone theorem ,

file:///E|/BOOK/BOOK6/INDEX_Z.HTM [19/1/2003 1:33:15]

Index (complete)

Up: Index - All

Index: (complete) #P-completeness, 0/1 knapsack problem 2-SAT 2/3 tree 3-SAT, k-optimal tours metric

-moves above-below test, abracadabra abstract data types abstract graph type academic institutions - licensing acceptance-rejection method Ackerman function acyclic subgraph Ada adaptive compression algorithms Adaptive Simulated Annealing (ASA) addition address book, TCS adjacency list, adjacency matrix, adjacent swaps advice - caveat aesthetically pleasing drawings aggregate range queries agrep file:///E|/BOOK/BOOK5/INDEX_AL.HTM (1 of 56) [19/1/2003 1:33:43]

Index (complete)

Aho-Corasick algorithm airline distance metric airline scheduling, algorist Algorist Technologies algorithm animation algorithm design algorithmic resources aligning DNA sequences alignment costs all-pairs shortest path,

,

alpha-beta pruning, alpha-shapes amortized analysis AMPL analog channel ancestor angle bisector animation - motion planning animation - sorting animation - systems annotated bibliography approximate nearest neighbor search, approximate string matching,

,

,

approximate string matching - related problems, approximate substring matching approximation algorithms, approximation scheme, arbitrage Arbitrary-Precision Arithmetic arbitrary-precision arithmetic - geometry arbitrary-precision arithmetic - related problems architectural models area computations - applications area computations - triangles area minimization file:///E|/BOOK/BOOK5/INDEX_AL.HTM (2 of 56) [19/1/2003 1:33:43]

Index (complete)

arm, robot around the world game Arrange, arrangement, , arrangement of objects arrangements of lines array array searching art gallery problems articulation vertex, artists steal ASA ASCII aspect ratio assembly language, assignment problem associative operation asymmetric longest path problem asymmetric TSPs, asymptotic analysis atom smashing attitude of the algorithm designer attribute attribute - graph augmenting path augmenting path methods authentication protocol automorphisms average average-case analysis average-case complexity AVL tree Avogadro's number awk Axiom axis-oriented rectangles,

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (3 of 56) [19/1/2003 1:33:43]

Index (complete)

axis-parallel planes B-tree, , , backpacker backsubstitution backtracking, , , , backtracking - animations

,

,

,

,

backtracking - applications, backtracking - bandwidth problem balanced search tree,

,

banded systems, bandersnatch problem bandwidth, bandwidth - matrix Bandwidth Reduction bandwidth reduction - backtracking bandwidth reduction - related problems bar codes base - arithmetic base - conversion base of logarithm Bellman-Ford algorithm, Berge's theorem best-case complexity Bible - searching the bibliographic databases biconnected components biconnected graphs, , big Oh notation bijection binary heap binary representation - subsets binary search,

,

binary search - applications, binary search - one-sided,

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (4 of 56) [19/1/2003 1:33:43]

,

Index (complete)

binary search tree, , , binary search tree - applications binary search tree - computational experience Bin Packing bin packing - applications, bin packing - knapsack problem bin packing - related problems, biocomputing biology bipartite graph bipartite graph recognition bipartite incidence structures bipartite matching,

,

,

bipartite matching - applications, bit-mapped images bit representation of graphs bit vector,

,

,

bit vector - applications, blind man's algorithm block - set partition blossoms board evaluation function bookshelves Boolean logic minimization, Boolean matrix multiplication borrowing Boruvka's algorithm boss's delight boundaries bounded height priority queue bounding boxes Boyer-Moore algorithm brainstorming branch-and-bound search, breadth-first search,

,

, ,

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (5 of 56) [19/1/2003 1:33:43]

Index (complete)

breadth-first search - applications bridge bridges of Königsberg Brook's theorem Brooks, Mel brush fire brute-force search bubblesort, bucketing techniques, , bucketing techniques - graphics bucket sort budget, fixed built-in random number generator buying fixed lots C++,

,

,

,

,

,

, , , , , , C++ templates cache Caesar shifts calculator, arithmetic Calendrical Calculations

, ,

, ,

, ,

, ,

call graph, canonically-labeled graphs canonical order, , CAP Carmichael numbers cars and tanks cartoons casino analysis casino poker catalog WWW site Catch-22 situation caveat CD-ROM, cdd

,

,

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (6 of 56) [19/1/2003 1:33:43]

,

,

,

,

,

,

,

,

,

,

,

,

Index (complete)

center vertex, , CGAL chain of matrices characters checksum chessboard coverage chess program, Chinese calendar Chinese postman problem Chinese remainder theorem Christofides heuristic chromatic index chromatic number chromatic polynomials cipher circle circuit analysis circuit board assembly circuit board placement - simulated annealing circuit layout circuit schematic diagrams circuit testing circular embeddings C language,

,

,

,

,

,

, , , , , , , , classification classification - nearest-neighbor classifiers - neural networks clauses clipping Clique

,

, ,

, ,

, ,

clique - applications, clique - definition clique - hardness proof clique - related problems clock closest pair heuristic file:///E|/BOOK/BOOK5/INDEX_AL.HTM (7 of 56) [19/1/2003 1:33:43]

, ,

, ,

, ,

, ,

, ,

,

,

,

,

,

,

Index (complete)

closest pair problem, closest point closure clothing - manufacturing cloudy days cluster clustered access cluster identification,

,

clustering, , co-NP coding theory coefficients cofactor method coin flip collapsing dense subgraphs Collected Algorithms of the ACM,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

, , , , , , collection coloring graphs color interchange combinatorial generation algorithms combinatorial geometry combinatorial problems, Combinatorica,

,

,

,

,

,

,

,

, , , , , , Commentz-Walter algorithm commercial implementations committee committee - congressional Common Lisp common substrings communication in circuits communications networks, comp.graphics.algorithms comp.theory

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (8 of 56) [19/1/2003 1:33:43]

,

,

,

,

,

,

,

,

,

,

,

,

Index (complete)

compaction comparisons - minimizing compiler compiler construction compiler optimization, compiler optimization - performance complement complement graph completion time - minimum complexity classes composite integer compress, compression compression - image computational biology computational complexity computational geometry computational number theory, computer algebra system, computer chess computer graphics computer graphics - applications, computer graphics - rendering computer vision concatenation - string concavities concavity elimination configurations configuration space conjugate gradient methods conjunctive normal form (CNF) connected components,

,

,

connected components - related problems, connected graph connectivity, , consensus sequences file:///E|/BOOK/BOOK5/INDEX_AL.HTM (9 of 56) [19/1/2003 1:33:43]

Index (complete)

consistent schedule Constrained and Unconstrained Optimization constrained and unconstrained optimization - related problems constrained Delaunay triangulation constrained optimization constrained optimization - related problems constraint elimination constraint satisfaction consulting services, container, context-free grammars Contig Assembly Program control systems - minimization convex decomposition, convex hull, Convex Hull convex hull - related problems, convex polygons convex polygons - intersection convex region convolution - polygon convolution - sequences cookbook cooling schedules coordinate transformations coplanar points copying a graph corporate ladder correctness - algorithm correlation function counterexample construction counting edges and vertices counting Eulerian cycles counting integer partitions counting linear extensions counting matchings

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (10 of 56) [19/1/2003 1:33:43]

Index (complete)

counting paths, counting spanning trees courses, lecture notes covering polygons with convex pieces covering set elements CPLEX Cramer's rule CRC critical path method crossing number crossings Cryptography cryptography - keys cryptography - related problems, CS CSA cubic regions currency speculation curve fitting Cuthill-McKee algorithm

,

cut set, cutting plane methods, cutting stock problem CWEB cycle - shortest cycle breaking cycle detection, cycle length cycle notation cycle structure of permutations cyclic-redundancy check (CRC) DAG DAG - longest path in DAG - shortest path in data abstraction database algorithms database application file:///E|/BOOK/BOOK5/INDEX_AL.HTM (11 of 56) [19/1/2003 1:33:43]

Index (complete)

database query optimization data compression Data Encryption Standard data filtering data records data structures, data structures - animations data transmission data validation Davenport-Schintzl sequences,

,

Davis-Putnam procedure, day of the week calculation deadlock de Bruijn sequence, debugging graph algorithms debugging parallel programs debugging randomized algorithms debugging time debugging tools decimal arithmetic decompose space decomposing polygons deconvolution decrease-key decryption Deep Blue defenestrate degeneracy degeneracy testing degenerate configuration degenerate system of equations degree, vertex, degree sequence degrees of freedom Delaunay triangulation, , Delaunay triangulation - applications

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (12 of 56) [19/1/2003 1:33:43]

Index (complete)

deletion from binary search tree deletions - text deliveries and pickups delivery routing Democrat/Republican identification De Morgan's laws dense graphs, , densest sphere packing dense subgraph depth-first search,

,

,

,

depth-first search - applications, depth-first search - backtracking derangement derivatives - automata derivatives - calculus DES descendent design process design rule checking determinant determinant - related problems Determinants and Permanents deterministic finite automata DFA diameter of a graph diameter of a point set Dictionaries

,

, ,

, ,

, ,

dictionaries - related problems, dictionary, , dictionary - applications dictionary - related problems dictionary - searching DIEHARD diff - how it works digital geometry digital signatures

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (13 of 56) [19/1/2003 1:33:43]

,

Index (complete)

Dijkstra's algorithm,

,

DIMACS, , DIMACS Challenge data DIMACS Implementation Challenge, Dinic's algorithm directed acyclic graph, , directed cycle directed graph directed graphs - automata directory file structures disclaimer

,

,

,

discrete event simulation, Discrete Fourier Transform, discrete mathematics software discussion section disjoint paths disjoint set union disjoint subsets disjunctive networks disjunctive normal form, disk access disk drives, dispatching emergency vehicles, dispersion problems distance graph distance metrics distinguishable elements distributed computation distribution sort, divide and conquer,

,

,

,

division, DNA DNA sequence comparisons DNA sequencing,

,

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (14 of 56) [19/1/2003 1:33:44]

,

,

Index (complete)

dominance orderings, DOS file names double-precision arithmetic, , Douglas-Plucker algorithm drawing graphs - related problems Drawing Graphs Nicely drawing puzzles Drawing Trees drawing trees - related problems, driving time minimization drug discovery DSATUR dual graph, duality, duality transformations duplicate elimination duplicate elimination - graphs duplicate elimination - permutations duplicate keys dynamic convex hulls dynamic data structures, dynamic graph algorithms dynamic Huffman codes dynamic programming,

,

,

,

,

,

,

dynamic programming - applications, , dynamic programming - initialization dynamic programming - shortest paths dynamic programming - space efficiency eavesdropper eccentricity of a graph economics - applications to edge/vertex connectivity - related problems, Edge and Vertex Connectivity edge chromatic number

,

edge coloring, file:///E|/BOOK/BOOK5/INDEX_AL.HTM (15 of 56) [19/1/2003 1:33:44]

Index (complete)

edge coloring - applications edge coloring - related problems, edge cover, , edge disjoint paths edge flipping operation edge labeled graphs edge length edge tour edit distance, Edmond's algorithm efficiency of algorithms eight-queens problem electrical engineers electronic circuit analysis electronic circuits Electronic Frontier Foundation element uniqueness problem, elimination ordering ellipsoid algorithm elliptic-curve method embeddings - planar Emde Boas priority queue empirical results, , empirical results - heuristics empirical results - how to do empirical results - string matching employees to jobs - matching empty circle - largest empty rectangle enclosing boxes enclosing disk enclosing rectangle encryption energy function energy minimization, English language, file:///E|/BOOK/BOOK5/INDEX_AL.HTM (16 of 56) [19/1/2003 1:33:44]

Index (complete)

English to French enumeration of spanning trees epsilon-moves equilateral triangle equivalence classes equivalence classes - automata states Erd•s-Gallai conditions error estimating closure sizes ethnic groups in Congress Euclid's algorithm Euclidean minimum spanning tree Euclidean traveling salesman Euler's formula Eulerian cycle - applications Eulerian cycle - line graphs Eulerian cycle - related problems, Eulerian Cycle / Chinese Postman Eulerian path evaluation function even-degree vertices even-length cycles event queue evolutionary tree exact cover problem exact string matching exam scheduling exercises,

,

,

,

,

exhaustive search, exhaustive search - application exhaustive search - empirical results exhaustive search - subsets expanded obstacles approach expander graphs expected-time, linear experimental analysis - set cover experimental graph theory

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (17 of 56) [19/1/2003 1:33:44]

Index (complete)

exponential-time algorithms, exponential distribution exponentiation, export restrictions external-memory sorting, external memory facets facility location, , Factoring and Primality Testing factoring and primality testing - related problems factoring integers - related problems factory location family tree, fan out minimization for networks FAQ file, , , farthest point Voronoi diagrams Fary's theorem faster computers fast Fourier transform fat cells fattening polygons feature sets Federal Sentencing Guidelines feedback edge set Feedback Edge/Vertex Set feedback edge/vertex set - related problems Fermat Fermat's theorem Ferrer's diagram FFT, FFTPACK fgrep Fibonacci heap, , , Fibonacci numbers FIFO file difference comparison file:///E|/BOOK/BOOK5/INDEX_AL.HTM (18 of 56) [19/1/2003 1:33:44]

Index (complete)

file directory trees file layout filtering outlying elements filtering signals final examination financial constraints find operation finite automata finite automata minimization finite element analysis Finite State Machine Minimization FIRE Engine firehouse first-fit - decreasing first in, first out fixed degree sequence graphs FLAP flat-earth model Fleury's algorithm flight crew scheduling floating-point arithmetic Floyd's algorithm, football program football scheduling Fortran,

,

,

,

,

,

,

,

,

,

,

, , , , , Fortune's algorithm

,

,

,

,

,

,

four-color problem, Fourier transform - applications Fourier transform - multiplication Fourier transform - related problems four Russians algorithm, fragment ordering fraud - tax freedom to hang yourself free space

,

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (19 of 56) [19/1/2003 1:33:44]

,

,

,

,

,

,

,

,

,

,

,

Index (complete)

free trees frequency distribution frequency domain friend-or-foe identification friendship graph, ftp - instructions function interpolation furniture moving furthest-point insertion heuristic furthest-site diagrams furthest-site Voronoi vertices future events game-tree search game-tree search - parallel games directory GAMS, gaps between primes garbage trucks Garey and Johnson Gates, William Gaussian distribution, Gaussian elimination, Genbank searching Generating Graphs Generating Partitions generating partitions - related problems, Generating Permutations

,

,

generating permutations - related problems, Generating Subsets generating subsets - applications generating subsets - related problems,

,

,

,

genetic algorithms, , , Genocop geographic information systems Geolab

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (20 of 56) [19/1/2003 1:33:44]

,

,

Index (complete)

geom.bib, geometric algorithms - animations geometric data structure geometric degeneracy geometric graphs geometric primitives - related problems geometric shortest path, geometric spanning tree geometric Steiner tree geometric traveling salesman problem geometric TSP GEOMPACK, Gettysburg Address Gibbs-Poole-Stockmeyer algorithm gift-wrapping algorithm Gilbert and Pollak conjecture Gingrich, Newt girth, glimpse global optimization Graffiti Graffiti - graphs of, Graham scan Grail graph algorithms, graph algorithms - animations graph algorithms - bandwidth problem GraphBase, , , graph complement

,

,

,

,

,

,

graph data structures, graph data structures - applications graph data structures - LEDA, graph density graph drawings - clutter GraphEd,

,

,

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (21 of 56) [19/1/2003 1:33:44]

,

,

,

,

Index (complete)

graph embedding graphical enumeration graphic partitions Graphics Gems graphics plotter graph isomorphism,

,

graph isomorphism - related problems, graph partition, Graph Partition graph partition - related problems, graph products graphs graph theory graph theory packages graph traversal GraphViz

,

Gray code, greatest common divisor greedy heuristic, , , , , , greedy heuristic - Huffman codes greedy heuristic - minimum spanning trees greedy heuristic - superstrings Gregorian calendar grid embeddings grid file, group - automorphism growth rates guarantees - importance of guarding art galleries Guide to Available Mathematical Software gzip, had-sex-with graph, half-space intersection Hamiltionian cycle - hypercube Hamiltonian cycle,

,

,

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (22 of 56) [19/1/2003 1:33:44]

Index (complete)

Hamiltonian Cycle Hamiltonian cycle - applications Hamiltonian cycle - counting Hamiltonian cycle - hardness proof Hamiltonian cycle - line graphs Hamiltonian cycle - related problems, Hamiltonian path Hamiltonian path - applications Hamming distance hardness of approximation hardware arithmetic hardware design applications hardware implementation hash function hash tables hash tables - computational experience hash tables - size Hausdorff distance heap heapsort, , , heard-of graph heart-lung machine heating ducts Hebrew calendar Hertel-Mehlhorn heuristic heuristics, , heuristics - empirical results hidden-surface elimination, hierarchical decomposition, hierarchical drawings hierarchical graph structures, hierarchy high-precision arithmetic - need for high-precision arithmetic - related problems, higher-dimensional data structures

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (23 of 56) [19/1/2003 1:33:44]

Index (complete)

higher-dimensional geometry, high school algebra high school cliques hill climbing historical objects

,

history, history - cryptography history - graph theory hitting set HIV virus homeomorphism homophones horizon Horner's rule, How to Solve It ht://Dig hub site Huffman codes Hull Human Genome Initiative Hungarian algorithm hypercube, hypergraph, , hyperlinks, WWW hyperplanes hypertext layout I_COLLIDE identical graphs IEEE Data Compression Conference image compression, image data image features image filtering image processing image segmentation image simplification

,

,

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (24 of 56) [19/1/2003 1:33:44]

Index (complete)

implementation challenge, DIMACS, implementation challenges, , , , , implementation complexity implementations, caveats implementation wanted implicit binary tree impress your friends algorithms in-circle test incidence matrices inconsistent linear equations increasing subsequences incremental algorithms incremental change methods incremental insertion algorithm incremental insertion algorithms - arrangements incremental insertion algorithms - coloring incremental insertion algorithms - graph drawing incremental insertion algorithms - sorting incremental insertion algorithms - suffix trees incremental insertion algorithms - TSP independent set, independent set - alternate formulations independent set - hardness proof independent set - related problems, , independent set - simulated annealing index - how to use

,

induced subgraph, induced subgraph isomorphism induction for algorithm design inequivalence of programs with assignments information retrieval information theory input/output graphics insertion into binary search tree insertions - text insertion sort,

,

,

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (25 of 56) [19/1/2003 1:33:44]

Index (complete)

inside/outside polygon instance - definition instance generator integer arithmetic integer factorization, integer partition, , integer programming

,

integer programming - applications, Integer programming - hardness proof integer programming - related problems integrality constraints interfering tasks interior-point methods Internal Revenue Service (IRS) Internet, , interpolation search intersection - halfspaces intersection - set Intersection Detection intersection detection - applications intersection detection - related problems, intersection point interview scheduling invariant - graph inverse Ackerman function inverse Fourier transform inverse matrix, inverse operations inversions Islamic calendar isolated vertex isomorphism isomorphism - graph isomorphism-complete iterative methods - linear systems jigsaw puzzle

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (26 of 56) [19/1/2003 1:33:44]

Index (complete)

job-shop scheduling job matching Job Scheduling Journal of Algorithms JPEG, Julian calendar Königsberg k-subset - applications k-subsets Karatsuba's algorithm Karazanov's algorithm Karmarkar's algorithm Karp-Rabin algorithm Kd-Trees, kd-trees - applications kd-trees - related problems, Kepler conjecture

,

Kernighan-Lin heuristic, key length, key search Kirchhoff's laws knapsack knapsack problem Knapsack Problem knapsack problem - applications knapsack problem - related problems knight's tour problem Knuth-Morris-Pratt algorithm Kolmogorov complexity Kruskal's algorithm, , , kth-order Voronoi diagrams kth-shortest path Kuratowski's theorem

,

labeled graphs, labeling maps label placement file:///E|/BOOK/BOOK5/INDEX_AL.HTM (27 of 56) [19/1/2003 1:33:44]

Index (complete)

labels language pattern matching LAPACK, large graphs - representation largest element last in, first out layered printed circuit boards Lazy adjacency matrix LCA - least common ancestor leap year least-squares curve fitting least common ancestor leaves - tree LEDA,

,

,

,

, , , , , left-right test left-to-right ordering

,

, ,

,

,

,

,

,

Lempel-Ziv algorithms, Lenstra's elliptic curve method lexicographic order, , , libraries licensing arrangements LIFO lifting-map construction line-point duality linear-time graph algorithms linear algebra, linear congruential generator linear constraint satisfaction linear extension linear interpolation search linear partitioning linear programming Linear Programming linear programming - models linear programming - related problems, file:///E|/BOOK/BOOK5/INDEX_AL.HTM (28 of 56) [19/1/2003 1:33:44]

,

,

,

,

,

,

,

,

,

,

,

,

Index (complete)

linear programming - relaxation linear programming - special cases line arrangements line graph, line intersection, line segment intersection line segment Voronoi diagram LINK, link distance, linked lists vs. arrays, LINPACK, , LISP list searching literate program locality of reference, local optima locations logarithms, logic minimization logic problems logic programming long division longest common prefix longest common substring, longest common substring - related problems, longest cycle, longest increasing subsequence, longest path, longest path - DAG long keys loop, lossless encodings lossy encodings

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (29 of 56) [19/1/2003 1:33:44]

Index (complete)

lottery problems Lotto problem low-degree spanning tree, low-dimensional linear programming lower bound, , , lower bound - range searching lower bound - sorting lower triangular matrix lp_solve LU-decomposition, lunar calendar LZW algorithm, machine-independent random number generator machine clock Macsyma mafia magnetic tape mail routing maintaining arrangements - related problems, Maintaining Line Arrangements makeg manufacturing applications, map labeling Maple map making marriage problems MAT, matching, matching - applications matching - dual to matching - number of perfect matching - related problems, matching shapes Mathematica,

,

,

,

, ,

, ,

, ,

,

,

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (30 of 56) [19/1/2003 1:33:44]

,

,

,

,

,

,

,

,

,

,

,

Index (complete)

, , , , , mathematical notation

,

,

mathematical programming, mathematical software - netlib matrix-tree theorem matrix bandwidth matrix compression matrix inversion, matrix multiplication Matrix Multiplication matrix multiplication - applications matrix multiplication - related problems matroids max-cut max-flow, min-cut theorem maxima maximal clique maximal matching maximum-cardinality matchings maximum acyclic subgraph maximum cut - simulated annealing maze, McDonald's restaurants MD5 mean Mechanical computers mechanical truss analysis medial-axis transform, Medial-Axis Transformation median - application Median and Selection medical residents to hospitals - matching memory accesses mems Menger's theorem mergesort,

,

,

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (31 of 56) [19/1/2003 1:33:44]

Index (complete)

merging subsets merging tapes mesh generation, Metaphone Metropolis algorithm middle square method millennium bug Miller-Rabin algorithm mindset minima minimax search minimizing automata minimum-change order minimum change order - subsets minimum cut minimum equivalent digraph minimum spanning tree, Minimum Spanning Tree

,

,

,

minimum spanning tree - applications, minimum spanning tree - drawing minimum spanning tree - related problems, minimum weight triangulation Minkowski metric

,

Minkowski sum, Minkowski sum - applications Minkowski sum - related problems, MIX assembly language mixed-integer programming mixed graphs mode, mode-switching modeling modeling algorithm problems modeling graph problems models of computation Modula-3 file:///E|/BOOK/BOOK5/INDEX_AL.HTM (32 of 56) [19/1/2003 1:33:44]

Index (complete)

modular arithmetic molecular docking molecular sequence data Mona Lisa, monotone decomposition monotone polygons Monte Carlo techniques, month and year morphing motion planning Motion Planning motion planning - related problems, , motion planning - shape simplification mountain climbing move to front rule, moving furniture MPEG, multicommodity flow multigraph multiple knapsacks multiple precision arithmetic multiple sequence alignment multiplication, multiplication, matrix multiplication algorithms multiset, musical scales name variations, recognizing naming concepts nanosecond national debt National Football League (NFL) National Security Agency (NSA) nauty, NC - Nick's class nearest-neighbor heuristic file:///E|/BOOK/BOOK5/INDEX_AL.HTM (33 of 56) [19/1/2003 1:33:44]

Index (complete)

nearest neighbor - related problems nearest neighbor graph, nearest neighbor search,

,

nearest neighbor search - related problems, negation negative-cost cycle negative cost edges NEOS, Netlib, , , , , , , , , network Network-Enabled Optimization System

,

network design, , network design - minimum spanning tree network flow, Network Flow network flow - applications, network flow - related problems,

,

,

,

network reliability, neural networks neural networks - classification neural networks - coloring, newsgroups next subset Nobel Prize, noisy channels noisy images, nonapproximability results noncrossing drawing nondeterministic automata nonEuclidean distance metrics nonnumerical problems nonorthogonal kd-tree nonself intersecting polygons nonuniform access file:///E|/BOOK/BOOK5/INDEX_AL.HTM (34 of 56) [19/1/2003 1:33:44]

,

Index (complete)

normal distribution notorious NP-complete problem NP, NP-completeness NP-completeness - definition of NP-completeness - proving NP-completeness - theory of NP-complete problem, , , NP-complete problem - bandwidth NP-complete problem - crossing number NP-complete problem - NFA minimization NP-complete problem - satisfiability NP-complete problem - set packing NP-complete problem - superstrings NP-complete problem - tetrahedralization NP-complete problem - tree drawing NP-complete problem - trie minimization NP-hard problems nuclear fission number field sieve number theory, numerical analysis numerical precision Numerical Recipes, numerical root finding numerical stability, O-notation objective function obstacle-filled rooms OCR octtree odd-degree vertices odd-length cycles, off-line problem oligonucleotide arrays on-line algorithm resources

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (35 of 56) [19/1/2003 1:33:44]

Index (complete)

on-line problem one-sided binary search, OpenGL graphics library operations research, optical character recognition, , , optical character recognition - system testing optimal binary search trees optimization problems ordered set ordering, order statistics organic graphs organ transplant orthogonal planes orthogonal polyline drawings orthogonal range query, outerplanar graphs outlying elements output-sensitive algorithms overdetermined linear systems overlap graph overpasses - highway Oxford English Dictionary P P-completeness p-tree packaging packaging applications packing vs. covering paging, pagoda pairing heap, palindrome paradigms of algorithms design parallel algorithms, parallel algorithms - graphs file:///E|/BOOK/BOOK5/INDEX_AL.HTM (36 of 56) [19/1/2003 1:33:44]

Index (complete)

parallel algorithms - visualization parallel lines parallel processor scheduling paranoia level parenthesization PARI, parse trees parsing partial key search partial order, partitioning automata states partitioning point sets partitioning polygons into convex pieces partitioning problems partition problem party affiliations Pascal, ,

, ,

, ,

, ,

, ,

, ,

,

,

,

,

,

password, patented algorithms path path generation - backtracking path planning paths - counting, Patricia trie pattern matching,

,

,

pattern recognition, pattern recognition - automata patterns Pat tree PDF-417 penalty functions perfect hashing perfect matching performance bottlenecks

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (37 of 56) [19/1/2003 1:33:44]

,

,

,

,

,

,

,

,

,

,

,

,

Index (complete)

performance guarantee performance in practice period periodicities perl permanent permutation, permutation comparisons permutation generation permutation generation - backtracking perpendicular bisector personality conflicts - avoiding PERT/CPM Petersen graph PGP, , phone company PHYLIP phylogenic tree, piano mover's problem Picasso, P., pieces of a graph pilots pink panther pivoting rules, pixel geometry, pLab planar drawings, planar drawings - related problems planar graph, planar graph - clique planar graph - coloring planar graph - instances planar graph - isomorphism Planarity Detection and Embedding planarity testing - related problems planar separators file:///E|/BOOK/BOOK5/INDEX_AL.HTM (38 of 56) [19/1/2003 1:33:44]

Index (complete)

planar subdivisions planar sweep algorithms plumbing point-spread function point distributions pointer manipulation point in polygon point location, point location - related problems, , , point robots points point set clusters Poisson distribution Polka polygonal data structure Polygon Partitioning polygon partitioning - related problems polygons polygon triangulation polyhedral simplification polyline graph drawings polynomial-time approximation scheme polynomial-time problems polynomial evaluation polynomial multiplication poor thin people popular keys porting code POSIT positions position tree potential function power diagrams power set powers of graphs Prüfer codes, precedence-constrainted scheduling

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (39 of 56) [19/1/2003 1:33:44]

Index (complete)

precedence constraints, precision preemptive scheduling prefix - string preflow-push methods preprocessing - graph algorithms presortedness measures Pretty Good Privacy previous subset PRF price-per-pound Prim's algorithm, primality testing, prime number prime number theorem principle of optimality printed circuit boards, printing a graph priority queues, priority queues - applications, , priority queues - arithmetic model priority queues - related problems problem - definition problem-specific algorithms problem descriptions problem instance

,

,

problem solving techniques, procedure call overhead producer/consumer sectors profit maximization Program Evaluation and Review Technique program libraries programming languages programming time, program structure Prolog file:///E|/BOOK/BOOK5/INDEX_AL.HTM (40 of 56) [19/1/2003 1:33:44]

Index (complete)

proof of correctness propagating consequences propositional logic protocol pruning - backtracking, pseudocode pseudorandom numbers psychic lotto prediction

,

public key cryptography,

,

Qhull, , , quadratic-sieve method quadratic programming quadtree quality triangulations questions queue, queue - applications quicksort, , , quicksort - applications rabbits radial embeddings radio stations radius of a graph radix sort, RAM RAM model of computation Random Access Machine (RAM) random generation - testing random graph theory, random graphs - generation randomization randomized algorithms, , randomized data structures

,

,

randomized incremental algorithms, randomized quicksort

,

,

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (41 of 56) [19/1/2003 1:33:44]

Index (complete)

randomized search - applications random number generation, , random number generation - related problems random permutations, random perturbations random sampling - applications random search tree random subset Ranger,

,

range search,

,

range search - related problems, ranked embedding ranking and unranking operations, ranking combinatorial objects ranking permutations ranking subsets RAPID rasterized images rational arithmetic ray shooting reachability problems rebalancing recommendations, caveat rectangle rectilinear Steiner tree recurrence relations recurrence relations - evaluation recursion - applications red-black tree

,

reduction, reduction - direction of reflex vertices region of influence regions regions formed by lines register allocation file:///E|/BOOK/BOOK5/INDEX_AL.HTM (42 of 56) [19/1/2003 1:33:44]

Index (complete)

regular expressions, relationship reliability, network repeated vertices representative selection Republican sex offenders resource allocation, resources - algorithm restricted growth function retrieval, reverse-search algorithms Right Stuff, The riots ensuing Rivest-Shamir-Adelman road network,

,

robot assembly, robot motion planning, , robust geometric computations Robust Geometric Primitives Roget's Thesaurus, rooted tree, root finding algorithms, rotating-calipers method rotation rotation - polygon roulette wheels round-off error round-off errors RSA-129

,

RSA algorithm, , rules of algorithm design run-length coding s-t connectivity safe cracker sequence

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (43 of 56) [19/1/2003 1:33:44]

Index (complete)

satisfiability, satisfiability - related problems, satisfying constraints sato scaling, scanner, OCR scattered subsequences scene interpolation scheduling, scheduling - precedence constraints scheduling - related problems, scheduling problems

,

Scheme, schoolhouse method sci.math scientific computing, Searching

,

searching - related problems, search space search time minimization - magnetic media search tree, secondary key secondary storage devices secure hashing function security, seed segmentation, segment intersection selection, , selection - subsets selection sort, self-intersecting polygons self-organizing list,

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (44 of 56) [19/1/2003 1:33:44]

Index (complete)

self-organizing tree, self-study textbook semi-exhaustive greedy algorithm semidefinite programming sentence structure separation problems separator theorems sequence sequencing by hybridization sequencing permutations sequential search, set set algorithms Set Cover, , set cover - applications set cover - exact set cover - related problems,

,

,

Set Data Structures, set data structures - applications set data structures - related problems Set Packing, set packing - related problems, set partition, sex offenders, Republican shape of a point set shape representation shapes Shape Similarity shape simplification shape simplification - applications, shellsort, Shifflett shift-register sequences shipping applications shipping problems file:///E|/BOOK/BOOK5/INDEX_AL.HTM (45 of 56) [19/1/2003 1:33:44]

Index (complete)

Shortest Common Superstring shortest common superstring - related problems, shortest cycle shortest path, Shortest Path

,

,

shortest path - applications, shortest path - definition shortest path - geometric, shortest path - related problems, shortest path matrix shotgun sequencing shuffling sieving devices - mechanical SIGACT sign - determinant sign - permutation signal processing signal propagation minimization

,

,

,

,

,

Sim++, SimPack, simple cycle simple graph simple polygon - construction simple polygons simplex method simplicial complex simplicity testing simplification envelopes Simplifying Polygons simplifying polygons - related problems simulated annealing, , , , simulated annealing - satisfiability simulated annealing - theory simulations simulations - accuracy sin, state of

,

,

,

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (46 of 56) [19/1/2003 1:33:45]

,

,

,

,

Index (complete)

sine functions single-precision numbers, single-source shortest path singular matrix, sinks - multiple sink vertex sites size of graph skeleton, skewed distribution Skiena, Len, skiing skinny triangles skip list slab method slack variables small edge weights smallest element smallest enclosing circle problem Smith Society smoothing, smoothness SNNS snow plows soap films software engineering software tools solar year Solving Linear Equations solving linear equations - related problems,

,

sorted array, sorted linked list, sorting, , sorting - applications sorting - animations file:///E|/BOOK/BOOK5/INDEX_AL.HTM (47 of 56) [19/1/2003 1:33:45]

Index (complete)

sorting - applications, sorting - cost of sorting - rationales for sorting - related problems, sorting - strings sound-alike strings

,

,

,

,

Soundex, sources - multiple source vertex space-efficient encodings space decomposition space minimization - digraphs space minimization - string matching spanning tree SPARE Parts sparse graph, , sparse matrices sparse matrices - compression sparse subset sparse systems sparsification spatial data structure special-purpose hardware speech recognition speedup - parallel spelling correction, sphere packing spikes Spinout puzzle spiral polygon

,

splay tree, SPLIB splicing cycles splines split-and-merge algorithm spreadsheet updates

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (48 of 56) [19/1/2003 1:33:45]

Index (complete)

spring embedding heuristics, square of a graph, square root of a graph square roots stable marriages, stable sorting stack, stack - applications stack size standard form Stanford GraphBase, , , star-shaped polygon decomposition state elimination, automata static tables statistical significance statistics steepest descent methods Steiner points Steiner ratio Steiner Tree Steiner tree - related problems Steiner vertices stock exchange stock picking Stony Brook Algorithm Repository Stony Brook class projects, straight-line graph drawings, Strassen's algorithm, , , strategy strength of a graph string string algorithms string algorithms - animations string data structures,

,

,

String Matching, file:///E|/BOOK/BOOK5/INDEX_AL.HTM (49 of 56) [19/1/2003 1:33:45]

Index (complete)

string matching - related problems, string overlaps strings strings - combinatorial strings - generating strongly-connected graphs strongly connected components

,

,

strongly connected graphs, Stuttgart Neural Network Simulator subgraph isomorphism subgraph isomorphism - applications subroutine call overhead, subset subset generation subset generation - backtracking subset sum problem substitution cipher substitutions, text substring matching, subtraction suffix array suffix trees,

,

suffix trees - applications, suffix trees - computational experience suffix trees - related problems, Suffix Trees and Arrays sunny days supercomputer superstrings - shortest common surface interpolation surface structures swap elements swapping sweepline algorithms, , symbolic computation symbolic set representation file:///E|/BOOK/BOOK5/INDEX_AL.HTM (50 of 56) [19/1/2003 1:33:45]

Index (complete)

Symbol Technologies symmetric difference symmetry detection symmetry removal tables tabu search tactics tail recursion take home lessons tape drive tax fraud taxonomy technical skills telephone books, telephone dialing

,

terrorist, test data testing planarity test pilots tetrahedralization text textbooks, text compression, Text Compression text compression - related problems,

,

text data structures, text processing algorithms text searching with errors thermodynamics thinning thinning - applications thinning - related problems, three-points-on-a-line tight bound time-series analysis time slot scheduling file:///E|/BOOK/BOOK5/INDEX_AL.HTM (51 of 56) [19/1/2003 1:33:45]

Index (complete)

Toeplitz matrices tool path optimization topological sorting, topological sorting - applications, topological sorting - related problems, topological sweep tour traffic light scheduling transition matrix transitive closure Transitive Closure and Reduction transitive reduction translation - polygon transmitter power transportation problems, transposition trapezoidal decomposition traveling salesman,

,

,

,

,

traveling salesman - applications, traveling salesman - approximation algorithms traveling salesman - decision problem traveling salesman - dynamic programming traveling salesman - related problems, , traveling salesman - simulated annealing Traveling Salesman Problem treap tree identification trees, trees - acyclic graphs trees - detection trees - drawings trees - generation trees - hard problem in trees - independent set trees - matching trees - partition file:///E|/BOOK/BOOK5/INDEX_AL.HTM (52 of 56) [19/1/2003 1:33:45]

Index (complete)

trial division Triangle triangle inequality, triangle refinement method triangle strips, triangulated surfaces Triangulation triangulation - applications, , triangulation - minimum weight

,

triangulation - related problems, triconnected components trie, trigram statistics TSP tsp_solve TSPLIB, turnpike reconstruction problem twenty questions two-coloring unbounded search, unconstrained optimization, , unconstrained optimization - related problems undirected graph uniform distribution, , union, set union-find data structure union-find data structure - applications union of polygons union of polygons - applications unit cube unit sphere universal set unknown data structures unlabeled graphs, unranking combinatorial objects file:///E|/BOOK/BOOK5/INDEX_AL.HTM (53 of 56) [19/1/2003 1:33:45]

Index (complete)

unranking permutations unranking subsets unsorted array unsorted list unweighted graphs - spanning trees upper bound upper triangular matrix Utah validation Vancouver Stock Exchange Vandermonde matrices vantage point tree variable elimination variable length encodings vector quantification vector sums Vertex Coloring, , , vertex coloring - applications vertex coloring - bipartite graphs vertex coloring - related problems,

,

vertex cover, vertex cover - approximation algorithm vertex cover - hardness proof, vertex cover - related problems,

,

vertex degree, vertex disjoint paths video - algorithm animation video compression, virtual memory, , virtual memory - algorithms virtual memory - performance virtual reality applications visibility graphs, Viterbi algorithm Vizing's theorem, file:///E|/BOOK/BOOK5/INDEX_AL.HTM (54 of 56) [19/1/2003 1:33:45]

Index (complete)

VLSI circuit layout, VLSI design problems volume computations, von Emde Boas queue von Neumann, J. Voronoi diagram Voronoi Diagrams Voronoi diagrams - nearest neighbor search Voronoi diagrams - related problems, walk-through

,

,

,

Waring's problem, Warshall's algorithm war story, water pipes wavelets

,

,

,

,

,

,

,

,

weakly-connected graphs, web weighted graphs, applications Winograd's algorithm wire length minimization wiring layout problems word ladders, worker assignment - scheduling world's record TSP worst-case complexity WWW site X-windows Xerox machines - scheduling XRLF XTango, , , , Young tableaux zero-knowledge proofs Zipf's law

,

,

,

,

,

,

zone theorem,

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (55 of 56) [19/1/2003 1:33:45]

,

,

,

,

,

,

,

,

,

Index (complete)

Algorithms Tue Jun 3 11:59:38 EDT 1997

file:///E|/BOOK/BOOK5/INDEX_AL.HTM (56 of 56) [19/1/2003 1:33:45]

1.4.4 Shortest Path

1.4.4 Shortest Path

INPUT

OUTPUT

Input Description: An edge-weighted graph G , with start vertex and end vertex t . Problem: Find the shortest path from to t in G .

Implementations ● ● ● ● ● ● ●

Goldberg's Network Optimization Codes (C) (rating 9) LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 7) Discrete Optimization Methods (Pascal) (rating 5) Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 4) Xtango and Polka Algorithm Animation Systems (C++) (rating 4) Combinatorica (Mathematica) (rating 3) The Stanford GraphBase (C) (rating 3)

file:///E|/WEBSITE/FILES/SHORPATH.HTM (1 of 2) [19/1/2003 1:33:53]

1.4.4 Shortest Path

Related Problems ● ● ● ● ● ● ● ●

Connected Components Graph Isomorphism Matrix Multiplication Motion Planning Network Flow Priority Queues Steiner Tree Transitive Closure and Reduction

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/SHORPATH.HTM (2 of 2) [19/1/2003 1:33:53]

1.2.5 Constrained and Unconstrained Optimization

1.2.5 Constrained and Unconstrained Optimization

INPUT

OUTPUT

Input Description: A function f(x_1,...,x_n) . Problem: What point p = (p_z,...,p_n) maximizes (or equivallently minimizes) the function f ?

Implementations ● ● ●

Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 8) Adaptive Simulated Annealing (C) (rating 6) Genocop -- Optimization via Genetic Algorithms (C) (rating 5)

Related Problems ●

Linear Programming

file:///E|/WEBSITE/FILES/UNCOTION.HTM (1 of 2) [19/1/2003 1:33:54]

1.2.5 Constrained and Unconstrained Optimization ● ●

Random Number Generation Satisfiability

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/UNCOTION.HTM (2 of 2) [19/1/2003 1:33:54]

1.6.4 Voronoi Diagrams

1.6.4 Voronoi Diagrams

INPUT

OUTPUT

Input Description: A set S of points p_1,...,p_n . Problem: Decompose the space into regions around each point, such that all the points in the region around p_i are closer to p_i than any other point in S .

Implementations ● ● ● ● ● ●

Fortune's 2D Voronoi diagram code (C) (rating 9) Qhull - higher dimensional convex hull program (C) (rating 7) LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 6) Joseph O'Rourke's Computational Geometry (C) (rating 4) The Stanford GraphBase (C) (rating 3) Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 3)

file:///E|/WEBSITE/FILES2/VORORAMS.HTM (1 of 2) [19/1/2003 1:33:56]

1.6.4 Voronoi Diagrams

Related Problems ● ● ● ● ●

Convex Hull Nearest Neighbor Search Point Location Medial-Axis Transformation Triangulation

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/VORORAMS.HTM (2 of 2) [19/1/2003 1:33:56]

1.4.7 Eulerian Cycle / Chinese Postman

1.4.7 Eulerian Cycle / Chinese Postman

INPUT

OUTPUT

Input Description: A graph G=(V,E) . Problem: Find the shortest tour of G visiting each edge at least once.

Implementations ● ●

Combinatorica (Mathematica) (rating 3) Nijenhuis and Wilf: Combinatorial Algorithms (FORTRAN) (rating 3)

Related Problems ● ●

Hamiltonian Cycle Matching

file:///E|/WEBSITE/FILES/EULEYCLE.HTM (1 of 2) [19/1/2003 1:33:58]

1.4.7 Eulerian Cycle / Chinese Postman

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/EULEYCLE.HTM (2 of 2) [19/1/2003 1:33:58]

Online Bibliographies

Online Bibliographies

There are many bibliographic sources available on the WWW, but the following I find indispensable: ●

●

●

●

Computer Science Bibliographies -- This is a collection of over 700,000 references to papers and technical reports in Computer Science, beneath a sophisticated search engine. While there is much duplication, this is my first stop whenever I need to look something up. The primary site is http://liinwww.ira.uka.de/bibliography/index.html , although several mirror sites are in operation around the world. All references are provided in bibtex format. Joel Seiferas' paper.lst -- References to over 55,000 papers and technical reports (at last count), mostly on algorithms and related areas of theoretical computer science. Each paper is reduced to a one-line format, which I find easy to grep through. It is available by anonymous ftp from ftp://ftp.cs.rochester.edu/pub/u/joel , but a local copy is also available. Strongly recommended. Geom.bib -- The complete bibliography on anything related to computational geometry, it references over 8,000 books, papers, and reports and includes detailed abstracts for many of them. It is available via anonymous ftp from ftp://ftp.cs.usask.ca/pub/geometry/geombib.tar.Z and a local copy is included. Grep-ing through geom.bib is an amazingly efficient way to find out about previous work without leaving your office. Compendium of Approximation Algorithms -- This compendium of approximation algorithms by Pierluigi Crescenzi and Viggo Kann of NP-hard optimization problems is available at http://www.nada.kth.se/nada/theory/problemlist.html, but a local copy is included. It is the place to look to find what is known about provably good heuristics for any given problem.

About the Book Send us Mail Go to Main Page

file:///E|/WEBSITE/BIBLIO/BIBLIO.HTM [19/1/2003 1:33:58]

About the Book -- The Algorithm Design Manual

About the Book

Most professional programmers are not well equipped to tackle algorithm design problems. The Algorithm Design Manual,written by Steven S. Skiena and published by Telos/Springer-Verlag is uniquely designed to provide access to combinatorial algorithms technology for computer professionals and students. This book is considerably different than other books on algorithms. Why? ●

●

●

●

●

We reduce the design process to a sequence of questions to ask about the problem at hand. This provides a concrete path to take the non-expert from an initial problem statement to a reasonable solution. To provide a better perspective on how algorithm problems arise in the real world, we include a collection of `war stories', tales from our experience on real problems. The moral of these stories is that algorithm design and analysis is not just theory, but an important tool to be pulled out and used as needed. Since the practical person is usually looking for a program more than an algorithm, we provide pointers to solid implementations whenever they are available. We have collected these implementations on an enclosed CD-ROM and at the book WWW site, http://www.cs.sunysb.edu/~algorith for easy retrieval. With these implementations available, the critical aspect in algorithm design becomes properly modeling your application, instead of becoming intimate with the details of the actual algorithm. This focus permeates the entire book. Since finding out what is known about an algorithmic problem can be a difficult task, we provide a catalog of the 75 most important algorithmic problems as a major component of this book. By browsing through this catalog, the reader can quickly identify what their problem called, what is known about it, and how they should proceed to solve it. As an aid in problem identification, we include a pair of `before' and `after' pictures for each problem, illustrating the required input and output specifications. The algorithm catalog spans numerical problems and data structures as well as graph, string, and geometric algorithms. For each problem in the catalog, we provide an honest and convincing motivation showing how it arises in applications. If we could not find such an application, then the

file:///E|/ABOUTBK.HTM (1 of 3) [19/1/2003 1:34:00]

About the Book -- The Algorithm Design Manual

problem doesn't appear in this book. Equally important is what we do not do in this book. We do not stress the mathematical analysis of algorithms, leaving most of the analysis as informal arguments. You will not find a single theorem anywhere in this book. But what is a manual without software? This book comes with a substantial electronic supplement, a ISO9660 compatible, multiplatform CD-ROM, which can be viewed using Netscape, Microsoft Explorer, or any other WWW browser. This CD-ROM contains: ●

●

●

A complete hypertext version of the full printed book. Indeed, the extensive cross-references within the text are best followed using the hypertext version. The source code and URLs for all cited implementations, mirroring the Algorithm Repository WWW site. Programs in C, C++, Fortran, and Pascal are included, providing an average of four different implementations for each algorithmic problem. Over 30 hours of audio lectures on the design and analysis of algorithms are provided, all keyed to on-line lecture notes. Following these lectures provides another approach to learning algorithm design techniques.

Together, this book covers material sufficient for a standard Introduction to Algorithms course. Its assumes the reader has completed the equivalent of a second programming course, typically titled Data Structures or Computer Science II . Special textbook oriented-features include: ●

●

●

●

In addition to standard pen-and-paper exercises, this book includes ``implementation challenges'' suitable for teams or individual students. These projects and the applied focus of the text can be used to provide a new laboratory focus to the traditional algorithms course. ``Take-home lessons'' at the beginning of each chapter emphasize the concepts to be gained from the chapter. This book stresses design over analysis. It is suitable for both traditional lecture courses, and the new ``active learning'' method, where the professor does not lecture instead guides student groups to solve real problems. The ``war stories'' provide a great introduction to the active learning method. A full set of lecture slides for teaching this course is available on the CD-ROM, keyed to unique on-line audio lectures covering a full semester algorithm course.

``I have not doubt that it will become a classic the day it is published. It has all the right ingredients: rich contents, friendly, personal language, subtle humor, the right references, and a plethora of pointers to resources.'' -- P. Takis Metaxas, Wellesley College. ``A major theme that runs through the book is that the most important technique to solve an algorithmic problem from the real world is to learn how to model the problem well. I did not believe this before; the

file:///E|/ABOUTBK.HTM (2 of 3) [19/1/2003 1:34:00]

About the Book -- The Algorithm Design Manual

book did an admirable job of convincing me that there is considerable truth in it.'' -- Giri Narasimhan, The University of Memphis. ``The questions on problem solving are good enough that they ought to be talked about in every programming class in the undergraduate curriculum.'' -- Ron Danielson, Santa Clara University. Check out the preface and table of contents for more information. You may order this book, and are encouraged to do so. You might also be interested in my previous book, Implementing Discrete Mathematics . Please leave your name and address to receive additional information about the book and notification of significant upgrades to this site when they occur.

If your WWW client does not support forms, please send an e-mail to [email protected] with your name, e-mail address, and mailing address for further information. First Name: Last Name : E-mail : Company Name : Number and Street: City : State : Zip : Phone : Reset

Register

Return to the home page If you have problems with this page, please send E-mail to: [email protected] file:///E|/ABOUTBK.HTM (3 of 3) [19/1/2003 1:34:00]

Copyright and Disclaimers

Copyright Notice

This electronic component package is protected by federal copyright law and international treaty. The copyright holder retains title to and ownership of the package. U.S. copyright law prohibits you from making any copy of the CD-ROM for any reason, without the written permission of Springer-Verlag, except that you may download and copy the contents of this CD-ROM disc for your own research, teaching, and communications use. Springer-Verlag or its designee has the right to audit your computer and electronic component usage to determine whether any unauthorized copies of this package have been made. For those programs and data contained on this CD-ROM and WWW site whereby the rights are owned by the original creators and source, any request for permission to use their software for commercial purposes must be directed to them. The printed book with which this electronic component is packaged may not be translated or copied in whole or in part without the written permission of Springer-Verlag except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software or by similar or dissimilar methodology now known or hereafter developed is forbidden. Springer-Verlag or the author makes no warranty of representation, either express or implied, with respect to this CD-ROM, WWW site, or book, including their quality, merchantability, or fitness for a particular purpose. In no event will Springer-Verlag, or the author be liable for direct, indirect, special, incidental, or consequential damages arising out of the use or inability to use the disc, WWW site, or book, even if Springer-Verlag or the author has been advised of the possibility of such damages.

About the Book Send us Mail Go to Main Page

file:///E|/WEBSITE/INFO/COPYRIGH.HTM [19/1/2003 1:34:01]

Image Mosaic

file:///E|/WEBSITE/IMAGES/TILES/TILES.HTM [19/1/2003 1:34:03]

CD-ROM Installation

Installation and Users Guide How do I start the CD-ROM? Load the CD-ROM into your drive. Viewing the contents of the CD-ROM requires you to have a properly installed WWW browser, such as the Netscape Navigator or Internet Explorer. A more detailed list of supported browsers is provided below. Open up the INDEX.HTM file on the CD-ROM with your browser, and then explore to your heart's content. The entire contents of the CD-ROM are accessible from this page.

How do I listen to the lectures? If you haven't already, you'll need to install the Shockwave for Director plug-in. Currently, Shockwave is supported only on Windows and Macintosh platforms, but keep checking Macromedia's website at www.macromedia.com for updated information about other platforms. Make sure that after installing the plug-in you restart your browser, or the plug-in will not be recognized. Copies of these plug-ins have been included on the CD-ROM. Click here to select the plug-in appropriate for your system.

What hardware and software do I need for this plugin? For Windows: Any modern PC computer will do, but a Pentium is recommended because of its floatingpoint unit. Supported browsers include the Netscape Navigator 2.02 and 3.01, Internet Explorer 3.0 and 3.01 Final Releases, (note: Internet Explorer for Windows can use the Shockwave plugins or the Shockwave ActiveX control, which can be automatically installed at http://www.macromedia.com), Attachmate's Emissary, Netmanage's WebSurfer, and America Online version 3.0 (note: no plug-in installation should be necessary for the AOL browser for Windows.) For Macintosh: Any PowerMac will do, but 68k Macs will require a hardware floating-point unit (math co-processor). Supported browsers include Netscape Navigator 2.02 and 3.01, Microsoft Internet Explorer 3.0 and 3.0a, Attachmate's Emissary, Netmanage's WebSurfer, and America Online version 3.0. Commercial versions of Netscape Navigator 3.x and Microsoft Internet Explorer may come with the appropriate plug-ins already installed. If not, copies of these plug-ins have been included on the CDROM. Click here to select the plug-in appropriate for your system.

To use the CD-ROM under Unix file:///E|/BROWSERS.HTM (1 of 3) [19/1/2003 1:34:04]

CD-ROM Installation

This CD-ROM is written as an ISO-9660 file system without Rock Ridge extensions. All filenames are written in upper case, and unless the file system is mounted correctly the HTML links will not be found. Unfortunately, different mount commands seem to be required for different flavors of Unix. Read your manual page and experiment. You must first create a mount point /cdrom. Let "sr0" be the SCSI ID for your CD-ROM drive (typically sr0 = "cdrom"). Possible mount commands which have worked on different systems include: mount -rt hsfs /dev/sr0 /cdrom mount -o norock,check=relaxed -r -t iso9660 /dev/sr0 /cdrom Once the CD-ROM is mounted, open /cdrom/index.html with your favorite browser. The entire contents of the CD-ROM are accessible from this page. Unfortunately, the audio lecture notes are not supported on Unix platforms.

About the Images Your browser likely has the option to automatically display or suppress in-line images. The mathematical symbols in the lecture notes and the textbook are represented by bit-mapped images, which might load slowly on certain systems. If so, disabling the automatic loading of images will result in better performance, although less readable text. To enable/disable viewing of these images when using Netscape Navigator, choose "Options" from the menu bar, and select "Auto Load Images". To view images on one page only, or to load images which were not properly loaded, click the "Images" button on the graphical toolbar.

About the Sound Sound files on this CD-ROM are recorded in Shockwave file format, and have the extension "SWA". Thus, the file "LEC19-2.SWA" contains sounds related to the second portion of lecture 19 in the lecture notes. Playing the audio requires a plug-in from Macromedia, which unfortunately is not available on Unix platforms. Click here to select the plug-in appropriate for your system. To listen to the sounds linked to the lecture notes you must: ●

Have an installed sound device configured for use on your system. If you have ever heard sound coming from your machine, this likely has already been done. If not, you will have to acquire and

file:///E|/BROWSERS.HTM (2 of 3) [19/1/2003 1:34:04]

CD-ROM Installation

●

install a sound device, such as Sound Blaster for PCs or the Macintosh standard sound system. Consult your sound device manual and the documentation for your operating system. Configure your browser to recognize sound files in an HTML document. In particular, the Netscape Navigator will have to launch an application to play the sound files. If Netscape attempts to save the audio file when you click on a sound link, this means that your sound application has not be properly installed.

About the Postscript Lecture Notes All the Lecture Notes are provided in Postscript format as well. To view Postscript files on Unix, use ghostview. Macintosh users: use the program DropPS 1.1.4, by Barebones Software. Windows users: use the program RoPS 32

file:///E|/BROWSERS.HTM (3 of 3) [19/1/2003 1:34:04]

Thanks!

Thanks! Four Stony Brook students made essential contributions to the production of this CD-ROM/WWW site. I'd like to thank them for their contributions by including their images for posterity. Thanks again, guys.

Ricky Bradley

file:///E|/WEBSITE/IMAGES/HELPERS/INDEX.HTM (1 of 4) [19/1/2003 1:34:07]

Thanks!

Frank Ruscica

file:///E|/WEBSITE/IMAGES/HELPERS/INDEX.HTM (2 of 4) [19/1/2003 1:34:07]

Thanks!

Dario Vlah

file:///E|/WEBSITE/IMAGES/HELPERS/INDEX.HTM (3 of 4) [19/1/2003 1:34:07]

Thanks!

Zhong Li Thanks also to Filip Bujanic, David Gerstl, Jim Klosowski, Kostis Sagonas, Kirsten Starcher, and Lei Zhao, who all made contributions at various stages of the project. Finally, I am grateful to acknowledge support from the Office of Naval Research and the National Science Foundation, which contributed towards this research.

file:///E|/WEBSITE/IMAGES/HELPERS/INDEX.HTM (4 of 4) [19/1/2003 1:34:07]

CD-ROM Installation

Installation and Users Guide How do I start the CD-ROM? Load the CD-ROM into your drive. Viewing the contents of the CD-ROM requires you to have a properly installed WWW browser, such as the Netscape Navigator or Internet Explorer. A more detailed list of supported browsers is provided below. Open up the INDEX.HTM file on the CD-ROM with your browser, and then explore to your heart's content. The entire contents of the CD-ROM are accessible from this page.

How do I listen to the lectures? If you haven't already, you'll need to install the Shockwave for Director plug-in. Currently, Shockwave is supported only on Windows and Macintosh platforms, but keep checking Macromedia's website at www.macromedia.com for updated information about other platforms. Make sure that after installing the plug-in you restart your browser, or the plug-in will not be recognized. Copies of these plug-ins have been included on the CD-ROM. Click here to select the plug-in appropriate for your system.

What hardware and software do I need for this plugin? For Windows: Any PC equipped with a floating-point unit will do, but a Pentium is recommended, especially since modern processors include a floating-point unit. Supported browsers include the Netscape Navigator 2.02 and 3.01, Internet Explorer 3.0 and 3.01 Final Releases, (note: Internet Explorer for Windows can use the Shockwave plugins or the Shockwave ActiveX control, which can be automatically installed at http://www.macromedia.com), Attachmate's Emissary, Netmanage's WebSurfer, and America Online version 3.0 (note: no plug-in installation should be necessary for the AOL browser for Windows.) For Macintosh: Any PowerMac will do, but 68k Macs will require a hardware floating-point unit (math co-processor). Supported browsers include Netscape Navigator 2.02 and 3.01, Microsoft Internet Explorer 3.0 and 3.0a, Attachmate's Emissary, Netmanage's WebSurfer, and America Online version 3.0. Commercial versions of Netscape Navigator 3.x and Microsoft Internet Explorer may come with the appropriate plug-ins already installed. If not, copies of these plug-ins have been included on the CDROM. Click here to select the plug-in appropriate for your system.

file:///E|/WEBSITE/INFO/BROWSERS.HTM (1 of 3) [19/1/2003 1:34:07]

CD-ROM Installation

To use the CD-ROM under Unix This CD-ROM is written as an ISO-9660 file system without Rock Ridge extensions. All filenames are written in upper case, and unless the file system is mounted correctly the HTML links will not be found. Unfortunately, different mount commands seem to be required for different flavors of Unix. Read your manual page. You must first create a mount point /cdrom. Let "sr0" be the SCSI ID for your CD-ROM drive (typically sr0 = "cdrom"). Possible mount commands which have worked on different systems include: mount -rt hsfs /dev/sr0 /cdrom mount -o norock,check=relaxed -r -t iso9660 /dev/sr0 /cdrom Once the CD-ROM is mounted, open /cdrom/index.html with your favorite browser. The entire contents of the CD-ROM are accessible from this page. Unfortunately, the audio lecture notes are not supported on Unix platforms.

About the Images Your browser likely has the option to automatically display or suppress in-line images. The mathematical symbols in the lecture notes and the textbook are represented by bit-mapped images, which might load slowly on certain systems. If so, disabling the automatic loading of images will result in better performance, although less readable text. To enable/disable viewing of these images when using Netscape Navigator, choose "Options" from the menu bar, and select "Auto Load Images". To view images on one page only, or to load images which were not properly loaded, click the "Images" button on the graphical toolbar.

About the Sound Sound files on this CD-ROM are recorded in Shockwave file format, and have the extension "SWA". Thus, the file "LEC19-2.SWA" contains sounds related to the second portion of lecture 19 in the lecture notes. Playing the audio requires a plug-in from Macromedia, which unfortunately is not available on Unix platforms. Click here to select the plug-in appropriate for your system. To listen to the sounds linked to the lecture notes you must:

file:///E|/WEBSITE/INFO/BROWSERS.HTM (2 of 3) [19/1/2003 1:34:07]

CD-ROM Installation ●

●

Have an installed sound device configured for use on your system. If you have ever heard sound coming from your machine, this likely has already been done. If not, you will have to acquire and install a sound device, such as Sound Blaster for PCs or the Macintosh standard sound system. Consult your sound device manual and the documentation for your operating system. Configure your browser to recognize sound files in an HTML document. In particular, the Netscape Navigator will have to launch an application to play the sound files. If Netscape attempts to save the audio file when you click on a sound link, this means that your sound application has not be properly installed.

About the Postscript Lecture Notes All the Lecture Notes are provided in Postscript format as well. To view Postscript files on Unix, use ghostview. Macintosh users: use the program DropPS 1.1.4, by Barebones Software. Windows-95 and NT users: use the program RoPS 32

file:///E|/WEBSITE/INFO/BROWSERS.HTM (3 of 3) [19/1/2003 1:34:07]

Binary Search in Action

Binary Search in Action Binary search is a fast algorithm for searching in a sorted array of keys. To look up a name in a telephone book with n names, you start by comparing the name that you want with the middle or (n/2) nd name, say onroe, Marilyn . Regardless of whether what you are looking someone before this middle name ( Dean, James ) or after it ( Presley, Elvis ), after this first comparison you can forever disregard one half of all the names in the book. The number of steps the algorithm takes equals the number of times we can halve n until only one name is left. Thus twenty comparisons suffice to find any name in the million-name Manhattan phone book! The power of binary search and logarithms is one of the most fundamental idea in the analysis of algorithms. This power becomes apparent if we imagine living in a world with only unsorted telephone books. The following animation of the first two stages of binary search is provided for your amusement. The full video tapes are also available.

file:///E|/WEBSITE/IMAGES/BINARY/INDEX.HTM (1 of 5) [19/1/2003 1:34:09]

Binary Search in Action

file:///E|/WEBSITE/IMAGES/BINARY/INDEX.HTM (2 of 5) [19/1/2003 1:34:09]

Binary Search in Action

file:///E|/WEBSITE/IMAGES/BINARY/INDEX.HTM (3 of 5) [19/1/2003 1:34:09]

Binary Search in Action

file:///E|/WEBSITE/IMAGES/BINARY/INDEX.HTM (4 of 5) [19/1/2003 1:34:09]

Binary Search in Action

file:///E|/WEBSITE/IMAGES/BINARY/INDEX.HTM (5 of 5) [19/1/2003 1:34:09]

About the Video Lectures

For Further Information In Spring 1996, my course on the Design and Analysis of Algorithms was offered as part of the EngiNet distance learning program at Stony Brook. As a result, all of my lectures were professionally video taped in a television studio. The picture and sound quality are excellent, certainly as good as possible given that they were recording me. EngiNet and the College of Engineering and Applied Science at Stony Brook have tentatively agreed to make these videotapes available to all interested parties, at a price of $20 per lecture, or $500 for the complete set of 25 lectures. Full lecture notes and audio sampled from the video tapes are available on line .

To obtain the video tapes, please send e-mail to [email protected] with the names of the desired lectures, your name, e-mail address, and mailing address.

file:///E|/WEBSITE/INFO/TAPES.HTM [19/1/2003 1:34:10]

Postscript version of the lecture notes

Lecture notes in Postscript format: These files contain postscript versions of the full lecture notes, formatted so that they are suitable for use as transparencies. Each page of the notes corresponds to an audio clip for the HTML versions of the notes. An index to these notes appears in the 300+ page file ALL. However, this file is not formatted as nicely as the individual lectures.

All

Lecture 1

Lecture 2

Lecture 3

Lecture 4

Lecture 5

Lecture 6

Lecture 7

Lecture 8

Lecture 9

Lecture 10

Lecture 11

Lecture 12

Lecture 13

Lecture 14

Lecture 15

Lecture 16

Lecture 17

Lecture 18

Lecture 19

Lecture 20

Lecture 21

Lecture 22

Lecture 23

To view Postscript files, Windows-95 and NT users should use the program RoPS 32. Macintosh users should use the program DropPS 1.1.4, by Barebones Software. On Unix, use the popular previewer ghostview.

file:///E|/LEC/LECTURES/PS/PS.HTM [19/1/2003 1:34:11]

Algorithm Repository -- Algorithms Courses

Algorithms Courses Instructors for combinatorial algorithms courses around the world are putting lecture notes and other interesting material on the WWW for their classes, but there is no reason the rest of the world shouldn't benefit as well. Below are the best course sites we have found to date. A similar collection of algorithm courses is being maintained by Kirk Pruhs at Univ. of Pittsburgh. ●

●

●

●

Gordon Royle -- Topics covered include graph algorithms, network flow, P=NP, travelling salesman, string processing, greedy algorithm, cryptography, and combinatorial search. On-line lecture notes are included. Steven Skiena -- Topic covered include sorting and searching, combinatorial search, dynamic programming, graph algorithms, and intractibility. On-line lecture notes are included. Guy Blelloch -- Parallel algorithms and programming. Included are animations of parallel algorithms, and information about the NESL parallel programming language they are written in. Algorithms in the Real World (CMU) -- An interesting collection of pointers and descriptive material on applications of algorithms in practice.

Feel free to look around. If you find something useful, then drop us a line to let us know what we did right. If you're looking for something that's not here yet, drop us a line and tell us that too. We're looking for as many good implementations as we can find -- so if you have an implementation for a specific problem, or know where one can be found, let us know and we'll see if we can't get it attached to the site. This algorithms repository is part of the ongoing algorithms-related research conducted here in the Department of Computer Science at the State University of New York at Stony Brook.

About the Book Send us Mail Go to Main Page

file:///E|/WEBSITE/INFO/COURSES.HTM [19/1/2003 1:34:11]

Lecture 1 - analyzing algorithms

Next: Lecture 2 - asymptotic Up: No Title Previous: No Title

Lecture 1 - analyzing algorithms Listen To Part 1-7

Lecture Schedule subject

topics

reading

Preliminaries

Analyzing algorithms

1-32

"

Asymptotic notation

32-37

"

Recurrence relations

53-64

Sorting

Heapsort

140-150

"

Quicksort

153-167

"

Linear Sorting

172-182

Searching

Data structures

200-215

"

Binary search trees

244-245

"

Red-Black trees:insertion

262-272

``

Red-Black trees:deletion

272-277

MIDTERM 1 Comb. Search

Backtracking

"

Elements of dynamic programming 301-314

"

Examples of dynamic programming 314-323

Graph Algorithms Data structures

465-477

for graphs "

Breadth/depth-first search

477-483

"

Topological Sort/Connectivity

485-493

"

Minimum Spanning Trees

498-510

"

Single-source shortest paths

514-532

"

All-pairs shortest paths

550-563

Intractability

P and NP

916-928

"

NP-completeness

929-939

"

NP-completeness proofs

939-951

MIDTERM 2

file:///E|/LEC/LECTUR16/NODE1.HTM (1 of 9) [19/1/2003 1:34:15]

Lecture 1 - analyzing algorithms

"

Further reductions

951-960

"

Approximation algorithms

964-974

"

Set cover / knapsack heuristics

974-983

FINAL EXAM Listen To Part 1-8

What Is An Algorithm? Algorithms are the ideas behind computer programs. An algorithm is the thing which stays the same whether the program is in Pascal running on a Cray in New York or is in BASIC running on a Macintosh in Kathmandu! To be interesting, an algorithm has to solve a general, specified problem. An algorithmic problem is specified by describing the set of instances it must work on and what desired properties the output must have.

Example: Sorting Input: A sequence of N numbers Output: the permutation (reordering) of the input sequence such as

.

We seek algorithms which are correct and efficient.

Correctness For any algorithm, we must prove that it always returns the desired output for all legal instances of the problem. For sorting, this means even if (1) the input is already sorted, or (2) it contains repeated elements.

Correctness is Not Obvious! The following problem arises often in manufacturing and transportation testing applications. Suppose you have a robot arm equipped with a tool, say a soldering iron. To enable the robot arm to do a soldering job, we must construct an ordering of the contact points, so the robot visits (and solders) the first contact point, then visits the second point, third, and so forth until the job is done. Since robots are expensive, we need to find the order which minimizes the time (ie. travel distance) it takes to assemble the circuit board.

file:///E|/LEC/LECTUR16/NODE1.HTM (2 of 9) [19/1/2003 1:34:15]

Lecture 1 - analyzing algorithms

You are given the job to program the robot arm. Give me an algorithm to find the best tour! Listen To Part 1-10

Nearest Neighbor Tour A very popular solution starts at some point etc. until done.

and then walks to its nearest neighbor

first, then repeats from

Pick and visit an initial point

i = 0

While there are still unvisited points

i = i+1

Let

be the closest unvisited point to

Visit

Return to

from

This algorithm is simple to understand and implement and very efficient. However, it is not correct!

file:///E|/LEC/LECTUR16/NODE1.HTM (3 of 9) [19/1/2003 1:34:15]

,

Lecture 1 - analyzing algorithms

Always starting from the leftmost point or any other point will not fix the problem. Listen To Part 1-11

Closest Pair Tour Always walking to the closest point is too restrictive, since that point might trap us into making moves we don't want.

Another idea would be to repeatedly connect the closest pair of points whose connection will not cause a cycle or a three-way branch to be formed, until we have a single chain with all the points in it.

Let n be the number of points in the set

For i=1 to n-1 do

For each pair of endpoints (x,y) of partial paths

If

then

,

Connect

by an edge

Connect the two endpoints by an edge.

Although it works correctly on the previous example, other data causes trouble:

file:///E|/LEC/LECTUR16/NODE1.HTM (4 of 9) [19/1/2003 1:34:15]

, d = dist(x,y)

Lecture 1 - analyzing algorithms

This algorithm is not correct! Listen To Part 1-12

A Correct Algorithm We could try all possible orderings of the points, then select the ordering which minimizes the total length:

For each of the n! permutations

If

of the n points

then

and

Return

Since all possible orderings are considered, we are guaranteed to end up with the shortest possible tour. Because it trys all n! permutations, it is extremely slow, much too slow to use when there are more than 10-20 points.

No efficient, correct algorithm exists for the traveling salesman problem, as we will see later. Listen To Part 1-13

Efficiency "Why not just use a supercomputer?" Supercomputers are for people too rich and too stupid to design efficient algorithms! A faster algorithm running on a slower computer will always win for sufficiently large instances, as we shall see.

file:///E|/LEC/LECTUR16/NODE1.HTM (5 of 9) [19/1/2003 1:34:15]

Lecture 1 - analyzing algorithms

Usually, problems don't have to get that large before the faster algorithm wins.

Expressing Algorithms We need some way to express the sequence of steps comprising an algorithm. In order of increasing precision, we have English, pseudocode, and real programming languages. Unfortunately, ease of expression moves in the reverse order. I prefer to describe the ideas of an algorithm in English, moving to pseudocode to clarify sufficiently tricky details of the algorithm. Listen To Part 1-14

The RAM Model Algorithms are the only important, durable, and original part of computer science because they can be studied in a machine and language independent way. The reason is that we will do all our design and analysis for the RAM model of computation: ● ●

●

Each "simple" operation (+, -, =, if, call) takes exactly 1 step. Loops and subroutine calls are not simple operations, but depend upon the size of the data and the contents of a subroutine. We do not want ``sort'' to be a single step operation. Each memory access takes exactly 1 step.

We measure the run time of an algorithm by counting the number of steps. This model is useful and accurate in the same sense as the flat-earth model (which is useful)! Listen To Part 1-15

Best, Worst, and Average-Case The worst case complexity of the algorithm is the function defined by the maximum number of steps taken on any instance of size n.

file:///E|/LEC/LECTUR16/NODE1.HTM (6 of 9) [19/1/2003 1:34:15]

Lecture 1 - analyzing algorithms

The best case complexity of the algorithm is the function defined by the minimum number of steps taken on any instance of size n. The average-case complexity of the algorithm is the function defined by an average number of steps taken on any instance of size n. Each of these complexities defines a numerical function - time vs. size!

Insertion Sort One way to sort an array of n elements is to start with proper position:

empty list, then successively insert new elements in the

At each stage, the inserted element leaves a sorted list, and after n insertions contains exactly the right elements. Thus the algorithm must be correct. But how efficient is it? Note that the run time changes with the permutation instance! (even for a fixed size problem) How does insertion sort do on sorted permutations? How about unsorted permutations?

Exact Analysis of Insertion Sort Count the number of times each line of pseudocode will be executed.

file:///E|/LEC/LECTUR16/NODE1.HTM (7 of 9) [19/1/2003 1:34:15]

Lecture 1 - analyzing algorithms

Line InsertionSort(A)

#Inst. #Exec.

1

for j:=2 to len. of A do

c1

n

2

key:=A[j]

c2

n-1

3

/* put A[j] into A[1..j-1] */ c3=0

4

i:=j-1

c4

n-1

5

while

do c5

tj

6

A[i+1]:= A[i]

c6

7

i := i-1

c7

8

A[i+1]:=key

c8

/

n-1

The for statement is executed (n-1)+1 times (why?) Within the for statement, "key:=A[j]" is executed n-1 times. Steps 5, 6, 7 are harder to count. Let

the number of elements that have to be slide right to insert the jth item. times.

Step 5 is executed .

Step 6 is

Add up the executed instructions for all pseudocode lines to get the run-time of the algorithm:

What are the

? They depend on the particular input.

Best Case If it's already sorted, all

's are 1.

Hence, the best case time is

where C and D are constants.

Worst Case If the input is sorted in descending order, we will have to slide all of the already-sorted elements, so

file:///E|/LEC/LECTUR16/NODE1.HTM (8 of 9) [19/1/2003 1:34:15]

, and step 5

Lecture 1 - analyzing algorithms

is executed

Next: Lecture 2 - asymptotic Up: No Title Previous: No Title Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR16/NODE1.HTM (9 of 9) [19/1/2003 1:34:15]

Lecture 2 - asymptotic notation

Next: Lecture 3 - recurrence Up: No Title Previous: Lecture 1 - analyzing

Lecture 2 - asymptotic notation Listen To Part 2-1 Problem 1.2-6: How can we modify almost any algorithm to have a good best-case running time?

To improve the best case, all we have to do it to be able to solve one instance of each size efficiently. We could modify our algorithm to first test whether the input is the special instance we know how to solve, and then output the canned answer. For sorting, we can check if the values are already ordered, and if so output them. For the traveling salesman, we can check if the points lie on a line, and if so output the points in that order. The supercomputer people pull this trick on the linpack benchmarks!

Because it is so easy to cheat with the best case running time, we usually don't rely too much about it. Because it is usually very hard to compute the average running time, since we must somehow average over all the instances, we usually strive to analyze the worst case running time. The worst case is usually fairly easy to analyze and often close to the average or real running time. Listen To Part 2-2

Exact Analysis is Hard! We have agreed that the best, worst, and average case complexity of an algorithm is a numerical function of the size of the instances.

file:///E|/LEC/LECTUR16/NODE2.HTM (1 of 10) [19/1/2003 1:34:21]

Lecture 2 - asymptotic notation

However, it is difficult to work with exactly because it is typically very complicated! Thus it is usually cleaner and easier to talk about upper and lower bounds of the function. This is where the dreaded big O notation comes in! Since running our algorithm on a machine which is twice as fast will effect the running times by a multiplicative constant of 2 - we are going to have to ignore constant factors anyway. Listen To Part 2-3

Names of Bounding Functions Now that we have clearly defined the complexity functions we are talking about, we can talk about upper and lower bounds on it: ● ● ●

g(n) = O(f(n)) means means means

Got it? C,

, and

is an upper bound on g(n). is a lower bound on g(n). is an upper bound on g(n) and

is a lower bound on g(n).

are all constants independent of n.

All of these definitions imply a constant values of n.

beyond which they are satisfied. We do not care about small

Listen To Part 2-4

O, , and

file:///E|/LEC/LECTUR16/NODE2.HTM (2 of 10) [19/1/2003 1:34:21]

Lecture 2 - asymptotic notation

The value of

shown is the minimum possible value; any greater value would also work.

(a) if there exist positive constants , , and f(n) always lies between and inclusive.

such that to the right of

, the value of

(b) f(n) = O(g(n)) if there are positive constants . always lies on or below

and c such that to the right of

, the value of f(n)

if there are positive constants (c) always lies on or above .

and c such that to the right of

, the value of f(n)

Asymptotic notation

are as well as we can practically deal with complexity functions.

Listen To Part 2-5

What does all this mean?

Think of the equality as meaning in the set of functions.

file:///E|/LEC/LECTUR16/NODE2.HTM (3 of 10) [19/1/2003 1:34:21]

Lecture 2 - asymptotic notation

Note that time complexity is every bit as well defined a function as function of time.

or you bank account as a

Listen To Part 2-6

Testing Dominance f(n) dominates g(n) if

, which is the same as saying g(n)=o(f(n)).

Note the little-oh - it means ``grows strictly slower than''. Knowing the dominance relation between common functions is important because we want algorithms whose time complexity is as low as possible in the hierarchy. If f(n) dominates g(n), f is much larger (ie. slower) than g. dominates

●

if a > b since

doesn't dominate

●

Complexity 10 n

since

20

30

40

50

60

0.00001 sec 0.00002 sec 0.00003 sec 0.00004 sec 0.00005 sec 0.00006 sec 0.0001 sec 0.0004 sec 0.0009 sec 0.016 sec

0.025 sec

0.036 sec

0.001 sec

0.008 sec

0.027 sec

0.064 sec

0.125 sec

0.216 sec

0.1 sec

3.2 sec

24.3 sec

1.7 min

5.2 min

13.0 min

0.001 sec

1.0 sec

17.9 min

12.7 days

35.7 years

366 cent

0.59 sec

58 min

6.5 years

3855 cent

cent

cent

Listen To Part 2-7

Logarithms It is important to understand deep in your bones what logarithms are and where they come from.

file:///E|/LEC/LECTUR16/NODE2.HTM (4 of 10) [19/1/2003 1:34:21]

Lecture 2 - asymptotic notation

A logarithm is simply an inverse exponential function. Saying .

is equivalent to saying that

Exponential functions, like the amount owed on a n year mortgage at an interest rate of per year, are functions which grow distressingly fast, as anyone who has tried to pay off a mortgage knows. Thus inverse exponential functions, ie. logarithms, grow refreshingly slowly. Binary search is an example of an algorithm. After each comparison, we can throw away half the possible number of keys. Thus twenty comparisons suffice to find any name in the million-name Manhattan phone book! If you have an algorithm which runs in large instances.

time, take it, because this is blindingly fast even on very

Listen To Part 2-8

Properties of Logarithms Recall the definition,

.

Asymptotically, the base of the log does not matter:

Thus,

, and note that

is just a constant.

Asymptotically, any polynomial function of n does not matter: Note that

since

, and

.

Any exponential dominates every polynomial. This is why we will seek to avoid exponential time file:///E|/LEC/LECTUR16/NODE2.HTM (5 of 10) [19/1/2003 1:34:21]

Lecture 2 - asymptotic notation

algorithms. Listen To Part 2-9

Federal Sentencing Guidelines 2F1.1. Fraud and Deceit; Forgery; Offenses Involving Altered or Counterfeit Instruments other than Counterfeit Bearer Obligations of the United States. (a) Base offense Level: 6 (b) Specific offense Characteristics (1) If the loss exceeded $2,000, increase the offense level as follows: Loss(Apply the Greatest) (A) $2,000 or less

Increase in Level no increase

(B) More than $2,000

add 1

(C) More than $5,000

add 2

(D) More than $10,000

add 3

(E) More than $20,000

add 4

(F) More than $40,000

add 5

(G) More than $70,000

add 6

(H) More than $120,000

add 7

(I) More than $200,000

add 8

(J) More than $350,000

add 9

(K) More than $500,000

add 10

(L) More than $800,000

add 11

(M) More than $1,500,000

add 12

(N) More than $2,500,000

add 13

(O) More than $5,000,000

add 14

(P) More than $10,000,000

add 15

(Q) More than $20,000,000

add 16

(R) More than $40,000,000

add 17

file:///E|/LEC/LECTUR16/NODE2.HTM (6 of 10) [19/1/2003 1:34:21]

Lecture 2 - asymptotic notation

(Q) More than $80,000,000

add 18

Listen To Part 2-10 The federal sentencing guidelines are designed to help judges be consistent in assigning punishment. The time-to-serve is a roughly linear function of the total level. However, notice that the increase in level as a function of the amount of money you steal grows logarithmically in the amount of money stolen. This very slow growth means it pays to commit one crime stealing a lot of money, rather than many small crimes adding up to the same amount of money, because the time to serve if you get caught is much less. The Moral: ``if you are gonna do the crime, make it worth the time!'' Listen To Part 2-11

Working with the Asymptotic Notation Suppose

and

.

What do we know about g'(n) = f(n)+g(n)? Adding the bounding constants shows

.

What do we know about g''(n) = f(n)-g(n)? Since the bounding constants don't necessary cancel,

We know nothing about the lower bounds on g'+g'' because we know nothing about lower bounds on f, g.

Suppose

and

.

What do we know about g'(n) = f(n)+g(n)? Adding the lower bounding constants shows What do we know about g''(n) = f(n)-g(n)? We know nothing about the lower bound of this! Listen To Part 2-12

The Complexity of Songs file:///E|/LEC/LECTUR16/NODE2.HTM (7 of 10) [19/1/2003 1:34:21]

.

Lecture 2 - asymptotic notation

Suppose we want to sing a song which lasts for n units of time. Since n can be large, we want to memorize songs which require only a small amount of brain space, i.e. memory. Let S(n) be the space complexity of a song which lasts for n units of time. The amount of space we need to store a song can be measured in either the words or characters needed to memorize it. Note that the number of characters is since every word in a song is at most 34 letters long - Supercalifragilisticexpialidocious! What bounds can we establish on S(n)? ●

●

S(n) = O(n), since in the worst case we must explicitly memorize every word we sing - ``The StarSpangled Banner'' , since we must know something about our song to sing it.

Listen To Part 2-13

The Refrain Most popular songs have a refrain, which is a block of text which gets repeated after each stanza in the song: Bye, bye Miss American pie Drove my chevy to the levy but the levy was dry Them good old boys were drinking whiskey and rye Singing this will be the day that I die. Refrains made a song easier to remember, since you memorize it once yet sing it O(n) times. But do they reduce the space complexity? Not according to the big oh. If

Then the space complexity is still O(n) since it is only halved (if the verse-size = refrain-size):

Listen To Part 2-14

file:///E|/LEC/LECTUR16/NODE2.HTM (8 of 10) [19/1/2003 1:34:21]

Lecture 2 - asymptotic notation

The k Days of Christmas To reduce S(n), we must structure the song differently. Consider ``The k Days of Christmas''. All one must memorize is: On the kth Day of Christmas, my true love gave to me, On the First Day of Christmas, my true love gave to me, a partridge in a pear tree But the time it takes to sing it is

If

, then

, so

.

Listen To Part 2-15

100 Bottles of Beer What do kids sing on really long car trips? n bottles of beer on the wall, n bottles of beer. You take one down and pass it around n-1 bottles of beer on the ball. All you must remember in this song is this template of size size for n depends on its value, but bits suffice. This for this song,

, and the current value of n. The storage

.

Is there a song which eliminates even the need to count? That's the way, uh-huh, uh-huh I like it, uh-huh, huh Reference: D. Knuth, `The Complexity of Songs', Comm. ACM, April 1984, pp.18-24 file:///E|/LEC/LECTUR16/NODE2.HTM (9 of 10) [19/1/2003 1:34:21]

Lecture 2 - asymptotic notation

Next: Lecture 3 - recurrence Up: No Title Previous: Lecture 1 - analyzing Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR16/NODE2.HTM (10 of 10) [19/1/2003 1:34:21]

Lecture 3 - recurrence relations

Next: Lecture 4 - heapsort Up: No Title Previous: Lecture 2 - asymptotic

Lecture 3 - recurrence relations Listen To Part 3-1 Problem 2.1-2: Show that for any real constants a and b, b > 0,

To show ●

●

, we must show O and

. Go back to the definition!

Big O - Must show that for all . When is this true? If all n > |a| since n+a < 2n, and raise both sides to the b. Big - Must show that for all . When is this true? If for all n > 3|a|/2 since n+a > n/2, and raise both sides to the b.

Note the need for absolute values. Listen To Part 3-2 Problem 2.1-4: (a) Is

?

(b) Is

?

(a) Is Is Yes, if

? ? for all n

(b) Is file:///E|/LEC/LECTUR17/NODE3.HTM (1 of 10) [19/1/2003 1:34:27]

, this is true for , this is true

Lecture 3 - recurrence relations

Is

?

note Is Is

? ?

No! Certainly for any constant c we can find an n such that this is not true. Listen To Part 3-3

Recurrence Relations Many algorithms, particularly divide and conquer algorithms, have time complexities which are naturally modeled by recurrence relations. A recurrence relation is an equation which is defined in terms of itself. Why are recurrences good things? 1. Many natural functions are easily expressed as recurrences:

2. It is often easy to find a recurrence as the solution of a counting problem. Solving the recurrence can be done for many special cases as we will see, although it is somewhat of an art. Listen To Part 3-4

Recursion is Mathematical Induction! In both, we have general and boundary conditions, with the general condition breaking the problem into smaller and smaller pieces.

file:///E|/LEC/LECTUR17/NODE3.HTM (2 of 10) [19/1/2003 1:34:27]

Lecture 3 - recurrence relations

The initial or boundary condition terminate the recursion. As we will see, induction provides a useful tool to solve recurrences - guess a solution and prove it by induction.

n 0 1 2 3 4 5 6 7 0 1 3 7 15 31 63 127 Guess what the solution is? Prove

by induction:

1. Show that the basis is true: . 2. Now assume true for 3. Using this assumption show:

.

Listen To Part 3-5

Solving Recurrences No general procedure for solving recurrence relations is known, which is why it is an art. My approach is:

Realize that linear, finite history, constant coefficient recurrences always can be solved Check out any combinatorics or differential equations book for a procedure. Consider

,

,

It has history = 2, degree = 1, and coefficients of 2 and 1. Thus it can be solved mechanically! Proceed: ●

Find the characteristic equation, eg.

file:///E|/LEC/LECTUR17/NODE3.HTM (3 of 10) [19/1/2003 1:34:27]

Lecture 3 - recurrence relations

● ● ●

Solve to get roots, which appear in the exponents. Take care of repeated roots and inhomogeneous parts. Find the constants to finish the job.

Systems like Mathematica and Maple have packages for doing this. Listen To Part 3-6

Guess a solution and prove by induction To guess the solution, play around with small values for insight. Note that you can do inductive proofs with the big-O's notations - just be sure you use it right. Example:

.

Show that

for large enough c and n. Assume that it is true for n/2, then

Starting with basis cases T(2)=4, T(3)=5, lets us complete the proof for

.

Listen To Part 3-7

Try backsubstituting until you know what is going on Also known as the iteration method. Plug the recurrence back into itself until you see a pattern. Example:

.

Try backsubstituting: file:///E|/LEC/LECTUR17/NODE3.HTM (4 of 10) [19/1/2003 1:34:27]

Lecture 3 - recurrence relations

The

term should now be obvious.

terms before we get to T(1), it doesn't hurt to sum them all since this is a Although there are only fast growing geometric series:

Listen To Part 3-8

Recursion Trees Drawing a picture of the backsubstitution process gives you a idea of what is going on. We must keep track of two things - (1) the size of the remaining argument to the recurrence, and (2) the additive stuff to be accumulated during this call. Example:

The remaining arguments are on the left, the additive terms on the right. Although this tree has height

, the total sum at each level decreases geometrically, so:

The recursion tree framework made this much easier to see than with algebraic backsubstitution. file:///E|/LEC/LECTUR17/NODE3.HTM (5 of 10) [19/1/2003 1:34:27]

Lecture 3 - recurrence relations

Listen To Part 3-9

See if you can use the Master theorem to provide an instant asymptotic solution The Master Theorem: Let and b>1 be constants, let f(n) be a function, and let T(n) be defined on the nonnegative integers by the recurrence

where we interpret n/b as

or

. Then T(n) can be bounded asymptotically as follows:

1. If for some constant , then 2. If for some constant 3. If sufficiently large n, then

, then . , and if

. for some constant c<1, and all

.

Listen To Part 3-10

Examples of the Master Theorem Which case of the Master Theorem applies? ●

T(n) = 4 T(n/2) + n Reading from the equation, a=4, b=2, and f(n) = n. Is

?

Yes, so case 1 applies and

.

●

Reading from the equation, a=4, b=2, and Is No, if

.

? , but it is true if

, so case 2 applies and

●

file:///E|/LEC/LECTUR17/NODE3.HTM (6 of 10) [19/1/2003 1:34:27]

.

Lecture 3 - recurrence relations

Reading from the equation, a=4, b=2, and Is

.

?

Yes, for

, so case 3 might apply.

Is

? , so there exists a c < 1 to satisfy the regularity condition, so case 3 applies and

Yes, for . Listen To Part 3-11

Why should the Master Theorem be true? Consider T(n) = a T(n/b) + f(n).

Suppose f(n) is small enough Say f(n)=0, ie. T(n) = a T(n/b). Then we have a recursion tree where the only contribution is at the leaves. There will be

levels, with

leaves at level l.

so long as f(n) is small enough that it is dwarfed by this, we have case 1 of the Master Theorem! Listen To Part 3-12 file:///E|/LEC/LECTUR17/NODE3.HTM (7 of 10) [19/1/2003 1:34:27]

Lecture 3 - recurrence relations

Suppose f(n) is large enough If we draw the recursion tree for T(n) = a T(n/b) + f(n).

If f(n) is a big enough function, the one top call can be bigger than the sum of all the little calls. Example:

. In fact this holds unless

!

In case 3 of the Master Theorem, the additive term dominates. In case 2, both parts contribute equally, which is why the log pops up. It is (usually) what we want to have happen in a divide and conquer algorithm. Listen To Part 3-13

Famous Algorithms and their Recurrence Matrix Multiplication The standard matrix multiplication algorithm for two

matrices is

.

Strassen discovered a divide-and-conquer algorithm which takes Since

dwarfs

, case 1 of the master theorem applies and

time. .

This has been ``improved'' by more and more complicated recurrences until the current best in Listen To Part 3-14

file:///E|/LEC/LECTUR17/NODE3.HTM (8 of 10) [19/1/2003 1:34:27]

.

Lecture 3 - recurrence relations

Polygon Triangulation Given a polygon in the plane, add diagonals so that each face is a triangle None of the diagonals are allowed to cross.

Triangulation is an important first step in many geometric algorithms. The simplest algorithm might be to try each pair of points and check if they see each other. If so, add the diagonal and recur on both halves, for a total of . However, Chazelle gave an algorithm which runs in

time. Since

, by case 1 of the Master Theorem, Chazelle's algorithm is linear, ie. T(n) = O(n).

Sorting The classic divide and conquer recurrence is Mergesort's T(n) = 2 T(n/2) + O(n), which divides the data into equal-sized halves and spends linear time merging the halves after they are sorted. Since

but not

, Case 2 of the Master Theorem applies and

. In case 2, the divide and merge steps balance out perfectly, as we usually hope for from a divide-andconquer algorithm.

Mergesort Animations Approaches to Algorithms Design Incremental Job is partly done - do a little more, repeat until done.

file:///E|/LEC/LECTUR17/NODE3.HTM (9 of 10) [19/1/2003 1:34:27]

Lecture 3 - recurrence relations

A good example of this approach is insertion sort

Divide-and-Conquer A recursive technique ● ●

●

Divide problem into sub-problems of the same kind. For subproblems that are really small (trivial), solve them directly. Else solve them recursively. (conquer) Combine subproblem solutions to solve the whole thing (combine)

A good example of this approach is Mergesort.

Next: Lecture 4 - heapsort Up: No Title Previous: Lecture 2 - asymptotic Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR17/NODE3.HTM (10 of 10) [19/1/2003 1:34:27]

Lecture 4 - heapsort

Next: Lecture 5 - quicksort Up: No Title Previous: Lecture 3 - recurrence

Lecture 4 - heapsort Listen To Part 4-1 4.2-2 Argue the solution to

is

by appealing to the recursion tree.

Draw the recursion tree.

How many levels does the tree have? This is equal to the longest path from the root to a leaf. The shortest path to a leaf occurs when we take the heavy branch each time. The height k is given by , meaning or . The longest path to a leaf occurs when we take the light branch each time. The height k is given by , meaning or . The problem asks to show that

, meaning we are looking for a lower bound

On any full level, the additive terms sums to n. There are

full levels. Thus

Listen To Part 4-2 4.2-4 Use iteration to solve T(n) = T(n-a) + T(a) + n, where

file:///E|/LEC/LECTUR17/NODE4.HTM (1 of 14) [19/1/2003 1:34:31]

is a constant.

Lecture 4 - heapsort

Note iteration is backsubstitution.

Listen To Part 4-3

Why don't CS profs ever stop talking about sorting?! 1. Computers spend more time sorting than anything else, historically 25% on mainframes. 2. Sorting is the best studied problem in computer science, with a variety of different algorithms known. 3. Most of the interesting ideas we will encounter in the course can be taught in the context of sorting, such as divide-and-conquer, randomized algorithms, and lower bounds. You should have seen most of the algorithms - we will concentrate on the analysis. Listen To Part 4-4

Applications of Sorting One reason why sorting is so important is that once a set of items is sorted, many other problems become easy.

Searching Binary search lets you test whether an item is in a dictionary in

time.

Speeding up searching is perhaps the most important application of sorting. file:///E|/LEC/LECTUR17/NODE4.HTM (2 of 14) [19/1/2003 1:34:31]

Lecture 4 - heapsort

Closest pair Given n numbers, find the pair which are closest to each other. Once the numbers are sorted, the closest pair will be next to each other in sorted order, so an O(n) linear scan completes the job. Listen To Part 4-5

Element uniqueness Given a set of n items, are they all unique or are there any duplicates? Sort them and do a linear scan to check all adjacent pairs. This is a special case of closest pair above.

Frequency distribution - Mode Given a set of n items, which element occurs the largest number of times? Sort them and do a linear scan to measure the length of all adjacent runs.

Median and Selection What is the kth largest item in the set? Once the keys are placed in sorted order in an array, the kth largest can be found in constant time by simply looking in the kth position of the array. Listen To Part 4-6

Convex hulls Given n points in two dimensions, find the smallest area polygon which contains them all.

file:///E|/LEC/LECTUR17/NODE4.HTM (3 of 14) [19/1/2003 1:34:31]

Lecture 4 - heapsort

The convex hull is like a rubber band stretched over the points. Convex hulls are the most important building block for more sophisticated geometric algorithms. Once you have the points sorted by x-coordinate, they can be inserted from left to right into the hull, since the rightmost point is always on the boundary. Without sorting the points, we would have to check whether the point is inside or outside the current hull. Adding a new rightmost point might cause others to be deleted.

Huffman codes If you are trying to minimize the amount of space a text file is taking up, it is silly to assign each letter the same length (ie. one byte) code. Example: e is more common than q, a is more common than z. If we were storing English text, we would want a and e to have shorter codes than q and z. To design the best possible code, the first and most important step is to sort the characters in order of frequency of use. Character Frequency Code f

5

1100

e

9

1101

c

12

100

b

13

101

d

16

111

a

45

0

file:///E|/LEC/LECTUR17/NODE4.HTM (4 of 14) [19/1/2003 1:34:31]

Lecture 4 - heapsort

Listen To Part 4-8

Selection Sort A simple

sorting algorithm is selection sort.

Sweep through all the elements to find the smallest item, then the smallest remaining item, etc. until the array is sorted.

Selection-sort(A)

for i = 1 to n

for j = i+1 to n

if (A[j] < A[i]) then swap(A[i],A[j])

It is clear this algorithm must be correct from an inductive argument, since the ith element is in its correct position. It is clear that this algorithm takes

time.

It is clear that the analysis of this algorithm cannot be improved because there will be n/2 iterations which will require at least n/2 comparisons each, so at least comparisons will be made. More careful analysis doubles this. Thus selection sort runs in

time.

Listen To Part 4-9

Binary Heaps A binary heap is defined to be a binary tree with a key in each node such that:

file:///E|/LEC/LECTUR17/NODE4.HTM (5 of 14) [19/1/2003 1:34:31]

Lecture 4 - heapsort

1. All leaves are on, at most, two adjacent levels. 2. All leaves on the lowest level occur to the left, and all levels except the lowest one are completely filled. 3. The key in root is all its children, and the left and right subtrees are again binary heaps. Conditions 1 and 2 specify shape of the tree, and condition 3 the labeling of the tree.

Listen To Part 4-10 The ancestor relation in a heap defines a partial order on its elements, which means it is reflexive, antisymmetric, and transitive. 1. Reflexive: x is an ancestor of itself. 2. Anti-symmetric: if x is an ancestor of y and y is an ancestor of x, then x=y. 3. Transitive: if x is an ancestor of y and y is an ancestor of z, x is an ancestor of z. Partial orders can be used to model heirarchies with incomplete information or equal-valued elements. One of my favorite games with my parents is fleshing out the partial order of ``big'' old-time movie stars.

The partial order defined by the heap structure is weaker than that of the total order, which explains 1. Why it is easier to build. 2. Why it is less useful than sorting (but still very important). Listen To Part 4-11

Constructing Heaps Heaps can be constructed incrementally, by inserting new elements into the left-most open spot in the array.

file:///E|/LEC/LECTUR17/NODE4.HTM (6 of 14) [19/1/2003 1:34:31]

Lecture 4 - heapsort

If the new element is greater than its parent, swap their positions and recur. Since at each step, we replace the root of a subtree by a larger one, we preserve the heap order. Since all but the last level is always filled, the height h of an n element heap is bounded because:

so

.

Doing n such insertions takes

, since the last n/2 insertions require

time each.

Listen To Part 4-12

Heapify The bottom up insertion algorithm gives a good way to build a heap, but Robert Floyd found a better way, using a merge procedure called heapify. Given two heaps and a fresh element, they can be merged into one by making the new one the root and trickling down.

Build-heap(A)

n = |A|

For

do

Heapify(A,i)

Heapify(A,i)

file:///E|/LEC/LECTUR17/NODE4.HTM (7 of 14) [19/1/2003 1:34:31]

Lecture 4 - heapsort

left = 2i

right = 2i+1

if

then

max = left

else max = i

if

and (A(right] > A[max]) then

max = right

if

then

swap(A[i],A[max])

Heapify(A,max)

Rough Analysis of Heapify Heapify on a subtree containing n nodes takes

The 2/3 comes from merging heaps whose levels differ by one. The last row could be exactly half filled. Besides, the asymptotic answer won't change so long the fraction is less than one. Solve the recurrence using the Master Theorem.

file:///E|/LEC/LECTUR17/NODE4.HTM (8 of 14) [19/1/2003 1:34:31]

Lecture 4 - heapsort

Let a = 1, b= 3/2 and f(n) = 1. Note that

, since

.

Thus Case 2 of the Master theorem applies.

The Master Theorem: Let integers by the recurrence

and b>1 be constants, let f(n) be a function, and let T(n) be defined on the nonnegative

where we interpret n/b to mean either

1. If 2. If 3. If

or

. Then T(n) can be bounded asymptotically as follows:

for some constant

, then

, then

.

.

for some constant

sufficiently large n, then

, and if

for some constant c<1, and all

.

Listen To Part 4-14

Exact Analysis of Heapify In fact, Heapify performs better than small.

, because most of the heaps we merge are extremely

In a full binary tree on n nodes, there are n/2 nodes which are leaves (i.e. height 0), n/4 nodes which are height 1, n/8 nodes which are height 2, ... In general, there are at most

nodes of height h, so the cost of building a heap is:

file:///E|/LEC/LECTUR17/NODE4.HTM (9 of 14) [19/1/2003 1:34:31]

Lecture 4 - heapsort

Since this sum is not quite a geometric series, we can't apply the usual identity to get the sum. But it should be clear that the series converges. Listen To Part 4-15

Proof of Convergence Series convergence is the ``free lunch'' of algorithm analysis. The identify for the sum of a geometric series is

If we take the derivative of both sides, ...

Multiplying both sides of the equation by x gives the identity we need:

Substituting x = 1/2 gives a sum of 2, so Build-heap uses at most 2n comparisons and thus linear time. Listen To Part 4-16

The Lessons of Heapsort, I "Are we doing a careful analysis? Might our algorithm be faster than it seems?" Typically in our analysis, we will say that since we are doing at most x operations of at most y time each, the total time is O(x y). However, if we overestimate too much, our bound may not be as tight as it should be!

file:///E|/LEC/LECTUR17/NODE4.HTM (10 of 14) [19/1/2003 1:34:31]

Lecture 4 - heapsort

Listen To Part 4-17

Heapsort Heapify can be used to construct a heap, using the observation that an isolated element forms a heap of size 1.

Heapsort(A)

Build-heap(A)

for i = n to 1 do

swap(A[1],A[i])

n = n - 1

Heapify(A,1)

If we construct our heap from bottom to top using Heapify, we do not have to do anything with the last n/2 elements. With the implicit tree defined by array positions, (i.e. the ith position is the parent of the 2ith and (2i+1)st positions) the leaves start out as heaps. Exchanging the maximum element with the last element and calling heapify repeatedly gives an sorting algorithm, named Heapsort.

Heapsort Animations The Lessons of Heapsort, II Always ask yourself, ``Can we use a different data structure?'' file:///E|/LEC/LECTUR17/NODE4.HTM (11 of 14) [19/1/2003 1:34:31]

Lecture 4 - heapsort

Selection sort scans throught the entire array, repeatedly finding the smallest remaining element.

For i = 1 to n

A:

Find the smallest of the first n-i+1 items.

B:

Pull it out of the array and put it first.

Using arrays or unsorted linked lists as the data structure, operation A takes O(n) time and operation B takes O(1). Using heaps, both of these operations can be done within a better tradeoff.

time, balancing the work and achieving

Listen To Part 4-19

Priority Queues A priority queue is a data structure on sets of keys supporting the following operations: ● ● ●

Insert(S, x) - insert x into set S Maximum(S) - return the largest key in S ExtractMax(S) - return and remove the largest key in S

These operations can be easily supported using a heap. ● ● ●

Insert - use the trickle up insertion in . Maximum - read the first element in the array in O(1). Extract-Max - delete first element, replace it with the last, decrement the element counter, then heapify in .

Listen To Part 4-20

Applications of Priority Queues

file:///E|/LEC/LECTUR17/NODE4.HTM (12 of 14) [19/1/2003 1:34:31]

Lecture 4 - heapsort

Heaps as stacks or queues

● ●

In a stack, push inserts a new item and pop removes the most recently pushed item. In a queue, enqueue inserts a new item and dequeue removes the least recently enqueued item.

Both stacks and queues can be simulated by using a heap, when we add a new time field to each item and order the heap according it this time field. ●

●

To simulate the stack, increment the time with each insertion and put the maximum on top of the heap. To simulate the queue, decrement the time with each insertion and put the maximum on top of the heap (or increment times and keep the minimum on top)

This simulation is not as efficient as a normal stack/queue implementation, but it is a cute demonstration of the flexibility of a priority queue.

Discrete Event Simulations In simulations of airports, parking lots, and jai-alai - priority queues can be used to maintain who goes next. The stack and queue orders are just special cases of orderings. In real life, certain people cut in line.

Sweepline Algorithms in Computational Geometry

In the priority queue, we will store the points we have not yet encountered, ordered by x coordinate. and push the line forward one stop at a time. Listen To Part 4-22 file:///E|/LEC/LECTUR17/NODE4.HTM (13 of 14) [19/1/2003 1:34:31]

Lecture 4 - heapsort

Greedy Algorithms In greedy algorithms, we always pick the next thing which locally maximizes our score. By placing all the things in a priority queue and pulling them off in order, we can improve performance over linear search or sorting, particularly if the weights change. Example: Sequential strips in triangulations.

Danny Heep

Next: Lecture 5 - quicksort Up: No Title Previous: Lecture 3 - recurrence Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR17/NODE4.HTM (14 of 14) [19/1/2003 1:34:31]

Lecture 5 - quicksort

Next: Lecture 6 - linear Up: No Title Previous: Lecture 4 - heapsort

Lecture 5 - quicksort Listen To Part 5-1 4-2 Find the missing integer from 0 to n using O(n) ``is bit[j] in A[i]'' queries.

Note - there are a total of

bits, so we are not allowed to read the entire input!

Also note, the problem is asking us to minimize the number of bits we read. We can spend as much time as we want doing other things provided we don't look at extra bits. How can we find the last bit of the missing integer? Ask all the n integers what their last bit is and see whether 0 or 1 is the bit which occurs less often than it is supposed to. That is the last bit of the missing integer! How can we determine the second-to-last bit? Ask the numbers which ended with the correct last bit! By analyzing the bit patterns of the numbers from 0 to n which end with this bit. By recurring on the remaining candidate numbers, we get the answer in T(n) = T(n/2) + n =O(n), by the Master Theorem. Listen To Part 5-2

Quicksort Although mergesort is merge.

, it is quite inconvenient for implementation with arrays, since we need space to

In practice, the fastest sorting algorithm is Quicksort, which uses partitioning as its main idea. Example: Pivot about 10. 17 12 6 19 23 8 5 10 - before 6 8 5 10 23 19 12 17 - after file:///E|/LEC/LECTUR17/NODE5.HTM (1 of 10) [19/1/2003 1:34:35]

Lecture 5 - quicksort

Partitioning places all the elements less than the pivot in the left part of the array, and all elements greater than the pivot in the right part of the array. The pivot fits in the slot between them. Note that the pivot element ends up in the correct place in the total order! Listen To Part 5-3

Partitioning the elements Once we have selected a pivot element, we can partition the array in one linear scan, by maintaining three sections of the array: < pivot, > pivot, and unexplored. Example: pivot about 10 | 17 12 6 19 23 8 5 | 10 | 5 12 6 19 23 8 | 17 5 | 12 6 19 23 8 | 17 5 | 8 6 19 23 | 12 17 5 8 | 6 19 23 | 12 17 5 8 6 | 19 23 | 12 17 5 8 6 | 23 | 19 12 17 5 8 6 ||23 19 12 17 5 8 6 10 19 12 17 23 As we scan from left to right, we move the left bound to the right when the element is less than the pivot, otherwise we swap it with the rightmost unexplored element and move the right bound one step closer to the left. Listen To Part 5-4 Since the partitioning step consists of at most n swaps, takes time linear in the number of keys. But what does it buy us? 1. The pivot element ends up in the position it retains in the final sorted order. 2. After a partitioning, no element flops to the other side of the pivot in the final sorted order.

file:///E|/LEC/LECTUR17/NODE5.HTM (2 of 10) [19/1/2003 1:34:35]

Lecture 5 - quicksort

Thus we can sort the elements to the left of the pivot and the right of the pivot independently! This gives us a recursive sorting algorithm, since we can use the partitioning approach to sort each subproblem. Listen To Part 5-5

Quicksort Animations Listen To Part 5-6

Pseudocode Sort(A)

Quicksort(A,1,n)

Quicksort(A, low, high)

if (low < high)

pivot-location = Partition(A,low,high)

Quicksort(A,low, pivot-location - 1)

Quicksort(A, pivot-location+1, high)

Partition(A,low,high)

pivot = A[low]

leftwall = low

file:///E|/LEC/LECTUR17/NODE5.HTM (3 of 10) [19/1/2003 1:34:35]

Lecture 5 - quicksort

for i = low+1 to high

if (A[i] < pivot) then

leftwall = leftwall+1

swap(A[i],A[leftwall])

swap(A[low],A[leftwall])

Listen To Part 5-7

Best Case for Quicksort Since each element ultimately ends up in the correct position, the algorithm correctly sorts. But how long does it take? The best case for divide-and-conquer algorithms comes when we split the input as evenly as possible. Thus in the best case, each subproblem is of size n/2. The partition step on each subproblem is linear in its size. Thus the total effort in partitioning the is O(n). size

problems of

The recursion tree for the best case looks like this:

levels of perfect partitions to get to single element The total partitioning on each level is O(n), and it take subproblems. When we are down to single elements, the problems are sorted. Thus the total time in the best case is . Listen To Part 5-8

file:///E|/LEC/LECTUR17/NODE5.HTM (4 of 10) [19/1/2003 1:34:35]

Lecture 5 - quicksort

Worst Case for Quicksort Suppose instead our pivot element splits the array as unequally as possible. Thus instead of n/2 elements in the smaller half, we get zero, meaning that the pivot element is the biggest or smallest element in the array.

Now we have n-1 levels, instead of elements to partition.

, for a worst case time of

, since the first n/2 levels each have

Thus the worst case time for Quicksort is worse than Heapsort or Mergesort. To justify its name, Quicksort had better be good in the average case. Showing this requires some fairly intricate analysis. The divide and conquer principle applies to real life. If you will break a job into pieces, it is best to make the pieces of equal size! Listen To Part 5-9

Intuition: The Average Case for Quicksort Suppose we pick the pivot element at random in an array of n keys.

Half the time, the pivot element will be from the center half of the sorted array. Whenever the pivot element is from positions n/4 to 3n/4, the larger remaining subarray contains at most 3n/4 elements. If we assume that the pivot element is always in this range, what is the maximum number of partitions we need to get from n elements down to 1 element?

file:///E|/LEC/LECTUR17/NODE5.HTM (5 of 10) [19/1/2003 1:34:35]

Lecture 5 - quicksort

Listen To Part 5-10

What have we shown? At most

levels of decent partitions suffices to sort an array of n elements.

But how often when we pick an arbitrary element as pivot will it generate a decent partition? Since any number ranked between n/4 and 3n/4 would make a decent pivot, we get one half the time on average. If we need levels of decent partitions to finish the job, and half of random partitions are decent, then on average the recursion tree to quicksort the array has levels.

Since O(n) work is done partitioning on each level, the average time is

.

More careful analysis shows that the expected number of comparisons is

.

Listen To Part 5-11

Average-Case Analysis of Quicksort To do a precise average-case analysis of quicksort, we formulate a recurrence given the exact expected time T(n):

Each possible pivot p is selected with equal probability. The number of comparisons needed to do the partition is n1.

file:///E|/LEC/LECTUR17/NODE5.HTM (6 of 10) [19/1/2003 1:34:35]

Lecture 5 - quicksort

We will need one useful fact about the Harmonic numbers

, namely

It is important to understand (1) where the recurrence relation comes from and (2) how the log comes out from the summation. The rest is just messy algebra. Listen To Part 5-12

rearranging the terms give us:

substituting

gives

We are really interested in A(n), so

Listen To Part 5-13

What is the Worst Case? file:///E|/LEC/LECTUR17/NODE5.HTM (7 of 10) [19/1/2003 1:34:35]

Lecture 5 - quicksort

The worst case for Quicksort depends upon how we select our partition or pivot element. If we always select either the first or last element of the subarray, the worst-case occurs when the input is already sorted! ABDFHJK BDFHJK DFHJK FHJK HJK JK K Having the worst case occur when they are sorted or almost sorted is very bad, since that is likely to be the case in certain applications. To eliminate this problem, pick a better pivot: 1. Use the middle element of the subarray as pivot. 2. Use a random element of the array as the pivot. 3. Perhaps best of all, take the median of three elements (first, last, middle) as the pivot. Why should we use median instead of the mean? Whichever of these three rules we use, the worst case remains a natural order it is much more difficult to occur.

. However, because the worst case is no longer

Listen To Part 5-14

Is Quicksort really faster than Heapsort? Since Heapsort is files.

and selection sort is

But how can we compare two notation, we can't!

, there is no debate about which will be better for decent-sized

algorithms to see which is faster? Using the RAM model and the big Oh

When Quicksort is implemented well, it is typically 2-3 times faster than mergesort or heapsort. The primary reason is that the operations in the innermost loop are simpler. The best way to see this is to implement both and experiment with different inputs. file:///E|/LEC/LECTUR17/NODE5.HTM (8 of 10) [19/1/2003 1:34:35]

Lecture 5 - quicksort

Since the difference between the two programs will be limited to a multiplicative constant factor, the details of how you program each algorithm will make a big difference. If you don't want to believe me when I say Quicksort is faster, I won't argue with you. It is a question whose solution lies outside the tools we are using. Listen To Part 5-15

Randomization Suppose you are writing a sorting program, to run on data given to you by your worst enemy. Quicksort is good on average, but bad on certain worst-case instances. If you used Quicksort, what kind of data would your enemy give you to run it on? Exactly the worst-case instance, to make you look bad. But instead of picking the median of three or the first element as pivot, suppose you picked the pivot element at random. Now your enemy cannot design a worst-case instance to give to you, because no matter which data they give you, you would have the same probability of picking a good pivot! Randomization is a very important and useful idea. By either picking a random pivot or scrambling the permutation before sorting it, we can say: ``With high probability, randomized quicksort runs in

time.''

Where before, all we could say is: ``If you give me random input data, quicksort runs in expected

time.''

Since the time bound how does not depend upon your input distribution, this means that unless we are extremely unlucky (as opposed to ill prepared or unpopular) we will certainly get good performance. Randomization is a general tool to improve algorithms with bad worst-case but good average-case complexity. The worst-case is still there, but we almost certainly won't see it.

Next: Lecture 6 - linear Up: No Title Previous: Lecture 4 - heapsort Algorithms file:///E|/LEC/LECTUR17/NODE5.HTM (9 of 10) [19/1/2003 1:34:35]

Lecture 5 - quicksort

Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR17/NODE5.HTM (10 of 10) [19/1/2003 1:34:35]

Lecture 6 - linear sorting

Next: Lecture 7 - elementary Up: No Title Previous: Lecture 5 - quicksort

Lecture 6 - linear sorting Listen To Part 6-1 7.1-2: Show that an n-element heap has height

.

Since it is balanced binary tree, the height of a heap is clearly answer.

, but the problem asks for an exact

The height is defined as the number of edges in the longest simple path from the root.

The number of nodes in a complete balanced binary tree of height h is Thus the height increases only when

, or in other words when

. is an integer.

Listen To Part 6-2 7.1-5 Is a reverse sorted array a heap?

In a heap, each element is greater than or equal to each of its descendants. In the array representation of a heap, the descendants of the ith element are the 2ith and (2i+1)th elements. If A is sorted in reverse order, then

implies that

file:///E|/LEC/LECTUR17/NODE6.HTM (1 of 7) [19/1/2003 1:34:38]

.

Lecture 6 - linear sorting

Since 2i > i and 2i+1 > i then

and

.

Thus by definition A is a heap! Listen To Part 6-3

Can we sort in better than

?

Any comparison-based sorting program can be thought of as defining a decision tree of possible executions. Running the same program twice on the same permutation causes it to do exactly the same thing, but running it on different permutations of the same data causes a different sequence of comparisons to be made on each.

Claim: the height of this decision tree is the worst-case complexity of sorting. Listen To Part 6-4 Once you believe this, a lower bound on the time complexity of sorting follows easily. Since any two different permutations of n elements requires a different sequence of steps to sort, there must be at least n! different paths from the root to leaves in the decision tree, ie. at least n! different leaves in the tree. Since only binary comparisons (less than or greater than) are used, the decision tree is a binary tree. Since a binary tree of height h has at most

leaves, we know

, or

.

, since the last n/2 terms of the product are each greater than n/2. By By inspection where e=2.718. Sterling's approximation, a better bound is

file:///E|/LEC/LECTUR17/NODE6.HTM (2 of 7) [19/1/2003 1:34:38]

Lecture 6 - linear sorting

Listen To Part 6-5

Non-Comparison-Based Sorting All the sorting algorithms we have seen assume binary comparisons as the basic primative, questions of the form ``is x before y?''. Suppose you were given a deck of playing cards to sort. Most likely you would set up 13 piles and put all cards with the same number in one pile. A 2 3 4 5 6 7 8 9 10 J Q K A 2 3 4 5 6 7 8 9 10 J Q K A 2 3 4 5 6 7 8 9 10 J Q K A 2 3 4 5 6 7 8 9 10 J Q K With only a constant number of cards left in each pile, you can use insertion sort to order by suite and concatenate everything together. If we could find the correct pile for each card in constant time, and each pile gets O(1) cards, this algorithm takes O(n) time. Listen To Part 6-6

Bucketsort Suppose we are sorting n numbers from 1 to m, where we know the numbers are approximately uniformly distributed. We can set up n buckets, each responsible for an interval of m/n numbers from 1 to m

Given an input number x, it belongs in bucket number

.

If we use an array of buckets, each item gets mapped to the right bucket in O(1) time. file:///E|/LEC/LECTUR17/NODE6.HTM (3 of 7) [19/1/2003 1:34:38]

Lecture 6 - linear sorting

With uniformly distributed keys, the expected number of items per bucket is 1. Thus sorting each bucket takes O(1) time! The total effort of bucketing, sorting buckets, and concatenating the sorted buckets together is O(n). What happened to our

lower bound!

Listen To Part 6-7 We can use bucketsort effectively whenever we understand the distribution of the data. However, bad things happen when we assume the wrong distribution. Suppose in the previous example all the keys happened to be 1. After the bucketing phase, we have:

We spent linear time distributing our items into buckets and learned nothing. Perhaps we could split the big bucket recursively, but it is not certain that we will ever win unless we understand the distribution. Problems like this are why we worry about the worst-case performance of algorithms! Such distribution techniques can be used on strings instead of just numbers. The buckets will correspond to letter ranges instead of just number ranges. The worst case ``shouldn't'' happen if we understand the distribution of our data. Listen To Part 6-8

Real World Distributions Consider the distribution of names in a telephone book. ● ● ●

Will there be a lot of Skiena's? Will there be a lot of Smith's? Will there be a lot of Shifflett's?

Either make sure you understand your data, or use a good worst-case or randomized algorithm! file:///E|/LEC/LECTUR17/NODE6.HTM (4 of 7) [19/1/2003 1:34:38]

Lecture 6 - linear sorting

The Shifflett's of Charlottesville For comparison, note that there are seven Shifflett's (of various spellings) in the 1000 page Manhattan telephone directory.

Listen To Part 6-10

Rules for Algorithm Design The secret to successful algorithm design, and problem solving in general, is to make sure you ask the right questions. Below, I give a possible series of questions for you to ask yourself as you try to solve difficult algorithm design problems: 1. Do I really understand the problem? 1. What exactly does the input consist of? 2. What exactly are the desired results or output? 3. Can I construct some examples small enough to solve by hand? What happens when I solve them? 4. Are you trying to solve a numerical problem? A graph algorithm problem? A geometric problem? A string problem? A set problem? Might your problem be formulated in more than one way? Which formulation seems easiest? 2. Can I find a simple algorithm for the problem? 1. Can I find the solve my problem exactly by searching all subsets or arrangements and picking the best one? 1. If so, why am I sure that this algorithm always gives the correct answer? 2. How do I measure the quality of a solution once I construct it?

file:///E|/LEC/LECTUR17/NODE6.HTM (5 of 7) [19/1/2003 1:34:38]

Lecture 6 - linear sorting

Listen To Part 6-11 3. Does this simple, slow solution run in polynomial or exponential time? 4. If I can't find a slow, guaranteed correct algorithm, am I sure that my problem is well defined enough to permit a solution? 2. Can I solve my problem by repeatedly trying some heuristic rule, like picking the biggest item first? The smallest item first? A random item first? 1. If so, on what types of inputs does this heuristic rule work well? Do these correspond to the types of inputs that might arise in the application? 2. On what types of inputs does this heuristic rule work badly? If no such examples can be found, can I show that in fact it always works well? 3. How fast does my heuristic rule come up with an answer? 3. Are there special cases of this problem I know how to solve exactly? 1. Can I solve it efficiently when I ignore some of the input parameters? 2. What happens when I set some of the input parameters to trivial values, such as 0 or 1? Listen To Part 6-12 3. Can I simplify the problem to create a problem I can solve efficiently? How simple do I have to make it? 4. If I can solve a certain special case, why can't this be generalized to a wider class of inputs? 4. Which of the standard algorithm design paradigms seem most relevant to the problem? 1. Is there a set of items which can be sorted by size or some key? Does this sorted order make it easier to find what might be the answer? 2. Is there a way to split the problem in two smaller problems, perhaps by doing a binary search, or a partition of the elements into big and small, or left and right? If so, does this suggest a divide-and-conquer algorithm? 3. Are there certain operations being repeatedly done on the same data, such as searching it for some element, or finding the largest/smallest remaining element? If so, can I use a data structure of speed up these queries, like hash tables or a heap/priority queue? 5. Am I still stumped? 1. Why don't I go back to the beginning of the list and work through the questions again? Do any of my answers from the first trip change on the second?

file:///E|/LEC/LECTUR17/NODE6.HTM (6 of 7) [19/1/2003 1:34:38]

Lecture 6 - linear sorting

Next: Lecture 7 - elementary Up: No Title Previous: Lecture 5 - quicksort Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR17/NODE6.HTM (7 of 7) [19/1/2003 1:34:38]

Lecture 7 - elementary data structures

Next: Lecture 8 - binary Up: No Title Previous: Lecture 6 - linear

Lecture 7 - elementary data structures Listen To Part 7-1 8.2-3 Argue that insertion sort is better than Quicksort for sorting checks

In the best case, Quicksort takes . Although using median-of-three turns the sorted permutation into the best case, we lose if insertion sort is better on the given data. In insertion sort, the cost of each insertion is the number of items which we have to jump over. In the check example, the expected number of moves per items is small, say c. We win if . Listen To Part 7-2 8.3-1 Why do we analyze the average-case performance of a randomized algorithm, instead of the worst-case?

In a randomized algorithm, the worst case is not a matter of the input but only of luck. Thus we want to know what kind of luck to expect. Every input we see is drawn from the uniform distribution. Listen To Part 7-3 8.3-2 How many calls are made to Random in randomized quicksort in the best and worst cases?

Each call to random occurs once in each call to partition. The number of partitions is

in any run of quicksort!!

There is some potential variation depending upon what you do with intervals of size 1 - do you call partition on intervals of size one? However, there is no asymptotic difference between best and worst case.

file:///E|/LEC/LECTUR17/NODE7.HTM (1 of 13) [19/1/2003 1:34:41]

Lecture 7 - elementary data structures

The reason - any binary tree with n leaves has n-1 internal nodes, each of which corresponds to a call to partition in the quicksort recursion tree. Listen To Part 7-4

Elementary Data Structures ``Mankind's progress is measured by the number of things we can do without thinking.'' Elementary data structures such as stacks, queues, lists, and heaps will be the ``of-the-shelf'' components we build our algorithm from. There are two aspects to any data structure: ● ●

The abstract operations which it supports. The implementation of these operations.

The fact that we can describe the behavior of our data structures in terms of abstract operations explains why we can use them without thinking, while the fact that we have different implementation of the same abstract operations enables us to optimize performance. Listen To Part 7-5

Stacks and Queues Sometimes, the order in which we retrieve data is independent of its content, being only a function of when it arrived. A stack supports last-in, first-out operations: push and pop. A queue supports first-in, first-out operations: enqueue and dequeue. A deque is a double ended queue and supports all four operations: push, pop, enqueue, dequeue. Lines in banks are based on queues, while food in my refrigerator is treated as a stack. Both can be used to traverse a tree, but the order is completely different.

file:///E|/LEC/LECTUR17/NODE7.HTM (2 of 13) [19/1/2003 1:34:41]

Lecture 7 - elementary data structures

Which order is better for WWW crawler robots? Listen To Part 7-6

Stack Implementation Although this implementation uses an array, a linked list would eliminate the need to declare the array size in advance.

STACK-EMPTY(S)

if top[S] = 0

then return TRUE

else return FALSE

PUSH(S, x)

POP(S)

file:///E|/LEC/LECTUR17/NODE7.HTM (3 of 13) [19/1/2003 1:34:41]

Lecture 7 - elementary data structures

if STACK-EMPTY(S)

then error ``underflow''

else

return S[top[S] + 1]

All are O(1) time operations. Listen To Part 7-7

Queue Implementation A circular queue implementation requires pointers to the head and tail elements, and wraps around to reuse array elements.

ENQUEUE(Q, x)

Q[tail[Q]]

x

if tail[Q] = length[Q]

then tail[Q]

1

else tail[Q]

tail[Q] + 1

file:///E|/LEC/LECTUR17/NODE7.HTM (4 of 13) [19/1/2003 1:34:41]

Lecture 7 - elementary data structures

DEQUEUE(Q)

x = Q[head[Q]]

if head[Q] =

length[Q]

then head[Q] = 1

else head[Q] = head[Q] + 1

return x

A list-based implementation would eliminate the possibility of overflow. All are O(1) time operations. Listen To Part 7-8

Dynamic Set Operations Perhaps the most important class of data structures maintain a set of items, indexed by keys. There are a variety of implementations of these dictionary operations, each of which yield different time bounds for various operations. ●

● ●

●

Search(S,k) - A query that, given a set S and a key value k, returns a pointer x to an element in S such that key[x] = k, or nil if no such element belongs to S. Insert(S,x) - A modifying operation that augments the set S with the element x. Delete(S,x) - Given a pointer x to an element in the set S, remove x from S. Observe we are given a pointer to an element x, not a key value. Min(S), Max(S) - Returns the element of the totally ordered set S which has the smallest (largest) key.

file:///E|/LEC/LECTUR17/NODE7.HTM (5 of 13) [19/1/2003 1:34:41]

Lecture 7 - elementary data structures ●

Next(S,x), Previous(S,x) - Given an element x whose key is from a totally ordered set S, returns the next largest (smallest) element in S, or NIL if x is the maximum (minimum) element.

Listen To Part 7-9

Pointer Based Implementation We can maintain a dictionary in either a singly or doubly linked list.

We gain extra flexibility on predecessor queries at a cost of doubling the number of pointers by using doublylinked lists. Since the extra big-Oh costs of doubly-linkly lists is zero, we will usually assume they are, although it might not be necessary. Singly linked to doubly-linked list is as a Conga line is to a Can-Can line.

Array Based Sets Unsorted Arrays ● ● ● ● ●

Search(S,k) - sequential search, O(n) Insert(S,x) - place in first empty spot, O(1) Delete(S,x) - copy nth item to the xth spot, O(1) Min(S,x), Max(S,x) - sequential search, O(n) Successor(S,x), Predecessor(S,x) - sequential search, O(n)

Listen To Part 7-10 Sorted Arrays ● ● ● ● ●

Search(S,k) - binary search, Insert(S,x) - search, then move to make space, O(n) Delete(S,x) - move to fill up the hole, O(n) Min(S,x), Max(S,x) - first or last element, O(1) Successor(S,x), Predecessor(S,x) - Add or subtract 1 from pointer, O(1)

What are the costs for a heap? file:///E|/LEC/LECTUR17/NODE7.HTM (6 of 13) [19/1/2003 1:34:42]

Lecture 7 - elementary data structures

Listen To Part 7-11

Unsorted List Implementation LIST-SEARCH(L, k)

x = head[L]

while x <> NIL and key[x] <> k

do x = next[x]

return x

Note: the while loop might require two lines in some programming languages.

LIST-INSERT(L, x)

next[x] = head[L]

if head[L] <> NIL

then prev[head[L]] = x

head[L] = x

file:///E|/LEC/LECTUR17/NODE7.HTM (7 of 13) [19/1/2003 1:34:42]

Lecture 7 - elementary data structures

prev[x] = NIL

LIST-DELETE(L, x)

if prev[x] <> NIL

then next[prev[x]] = next[x]

else head[L] = next[x]

if next[x] <> NIL

then prev[next[x]] = prev[x]

Sentinels Boundary conditions can be eliminated using a sentinel element which doesn't go away.

LIST-SEARCH'(L, k)

x = next[nil[L]]

while x <> NIL[L] and key[x] <> k

do x = next[x]

file:///E|/LEC/LECTUR17/NODE7.HTM (8 of 13) [19/1/2003 1:34:42]

Lecture 7 - elementary data structures

return x

LIST-INSERT'(L, x)

next[x] = next[nil[L]]

prev[next[nil[L]]] = x

next[nil[L]] = x

prev[x] = NIL[L]

LIST-DELETE'(L, x)

next[prev[x]] <> next[x]

next[prev[x]] = prev[x]

Listen To Part 7-13

Hash Tables Hash tables are a very practical way to maintain a dictionary. As with bucket sort, it assumes we know that the distribution of keys is fairly well-behaved. The idea is simply that looking an item up in an array is mathematical function which maps keys to integers.

once you have its index. A hash function is a

In bucket sort, our hash function mapped the key to a bucket based on the first letters of the key. ``Collisions'' were the set of keys mapped to the same bucket.

file:///E|/LEC/LECTUR17/NODE7.HTM (9 of 13) [19/1/2003 1:34:42]

Lecture 7 - elementary data structures

If the keys were uniformly distributed, then each bucket contains very few keys! The resulting short lists were easily sorted, and could just as easily be searched!

Listen To Part 7-14

Hash Functions It is the job of the hash function to map keys to integers. A good hash function: 1. Is cheap to evaluate 2. Tends to use all positions from with uniform frequency. 3. Tends to put similar keys in different parts of the tables (Remember the Shifletts!!) The first step is usually to map the key to a big integer, for example

This large number must be reduced to an integer whose size is between 1 and the size of our hash table. One way is by the high bits.

, where M is best a large prime not too close to

This works on the same principle as a roulette wheel! Listen To Part 7-15

Good and Bad Hash functions The first three digits of the Social Security Number

file:///E|/LEC/LECTUR17/NODE7.HTM (10 of 13) [19/1/2003 1:34:42]

, which would just mask off

Lecture 7 - elementary data structures

The last three digits of the Social Security Number

Listen To Part 7-16

The Birthday Paradox No matter how good our hash function is, we had better be prepared for collisions, because of the birthday paradox.

The probability of there being no collisions after n insertions into an m-element table is

When m = 366, this probability sinks below 1/2 when N = 23 and to almost 0 when

file:///E|/LEC/LECTUR17/NODE7.HTM (11 of 13) [19/1/2003 1:34:42]

.

Lecture 7 - elementary data structures

Listen To Part 7-17

Collision Resolution by Chaining The easiest approach is to let each element in the hash table be a pointer to a list of keys.

Insertion, deletion, and query reduce to the problem in linked lists. If the n keys are distributed uniformly in a table of size m/n, each operation takes O(m/n) time. Chaining is easy, but devotes a considerable amount of memory to pointers, which could be used to make the table larger. Still, it is my preferred method. Listen To Part 7-18

Open Addressing We can dispense with all these pointers by using an implicit reference derived from a simple function:

If the space we want to use is filled, we can examine the remaining locations: 1. Sequentially 2. Quadratically 3. Linearly The reason for using a more complicated science is to avoid long runs from similarly hashed keys. Deletion in an open addressing scheme is ugly, since removing one element can break a chain of insertions, making some elements inaccessible. Listen To Part 7-19

Performance on Set Operations With either chaining or open addressing:

file:///E|/LEC/LECTUR17/NODE7.HTM (12 of 13) [19/1/2003 1:34:42]

Lecture 7 - elementary data structures ● ● ● ●

Search - O(1) expected, O(n) worst case Insert - O(1) expected, O(n) worst case Delete - O(1) expected, O(n) worst case Min, Max and Predecessor, Successor

expected and worst case

Pragmatically, a hash table is often the best data structure to maintain a dictionary. However, we will not use it much in proving the efficiency of our algorithms, since the worst-case time is unpredictable. The best worst-case bounds come from balanced binary trees, such as red-black trees.

Next: Lecture 8 - binary Up: No Title Previous: Lecture 6 - linear Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR17/NODE7.HTM (13 of 13) [19/1/2003 1:34:42]

Lecture 8 - binary trees

Next: Lecture 9 - catch Up: No Title Previous: Lecture 7 - elementary

Lecture 8 - binary trees Listen To Part 8-1 9.1-3 Show that there is no sorting algorithm which sorts at least

Think of the decision tree which can do this.

instances in O(n) time.

What is the shortest tree with

leaves?

Moral: there cannot be too many good cases for any sorting algorithm! Listen To Part 8-2 9.1-4 Show that the

lower bound for sorting still holds with ternary comparisons.

The maximum number of leaves in a tree of height h is

,

So it goes for any constant base. Listen To Part 8-3

Binary Search Trees ``I think that I shall never see a poem as lovely as a tree Poem's are wrote by fools like me but only G-d can make a tree ``

file:///E|/LEC/LECTUR17/NODE8.HTM (1 of 13) [19/1/2003 1:34:46]

Lecture 8 - binary trees

- Joyce Kilmer Binary search trees provide a data structure which efficiently supports all six dictionary operations. A binary tree is a rooted tree where each node contains at most two children. Each child can be identified as either a left or right child.

A binary tree can be implemented where each node has left and right pointer fields, an (optional) parent pointer, and a data field. Listen To Part 8-4

Binary Search Trees A binary search tree labels each node in a binary tree with a single key such that for any node x, and nodes in the left subtree and all nodes in the right subtree of x have key's . of x have keys

Left: A binary search tree. Right: A heap but not a binary search tree. The search tree labeling enables us to find where any key is. Start at the root - if that is not the one we want, search either left or right depending upon whether what we want is or then the root. Listen To Part 8-5

Searching in a Binary Tree Dictionary search operations are easy in binary trees ...

TREE-SEARCH(x, k) file:///E|/LEC/LECTUR17/NODE8.HTM (2 of 13) [19/1/2003 1:34:46]

Lecture 8 - binary trees

if (x = NIL) and (k = key[x])

then return x

if (k < key[x])

then return TREE-SEARCH(left[x],k)

else return TREE-SEARCH(right[x],k)

The algorithm works because both the left and right subtrees of a binary search tree are binary search trees - recursive structure, recursive algorithm. This takes time proportional to the height of the tree, O(h). Listen To Part 8-6

Maximum and Minimum Where are the maximum and minimum elements in a binary tree?

TREE-MAXIMUM(X)

while

do x = right[x]

file:///E|/LEC/LECTUR17/NODE8.HTM (3 of 13) [19/1/2003 1:34:46]

Lecture 8 - binary trees

return x

TREE-MINIMUM(x)

while

do x = left[x]

return x

Both take time proportional to the height of the tree, O(h). Listen To Part 8-7

Where is the predecessor? Where is the predecessor of a node in a tree, assuming all keys are distinct?

If X has two children, its predecessor is the maximum value in its left subtree and its successor the minimum value in its right subtree. Listen To Part 8-8

What if a node doesn't have children?

file:///E|/LEC/LECTUR17/NODE8.HTM (4 of 13) [19/1/2003 1:34:46]

Lecture 8 - binary trees

If it does not have a left child, a node's predecessor is its first left ancestor. The proof of correctness comes from looking at the in-order traversal of the tree.

Tree-Successor(x)

if

then return Tree-Minimum(right[x])

while

and (x = right[y])

do

return y

Tree predecessor/successor both run in time proportional to the height of the tree. Listen To Part 8-9

In-Order Traversal

file:///E|/LEC/LECTUR17/NODE8.HTM (5 of 13) [19/1/2003 1:34:46]

Lecture 8 - binary trees

Inorder-Tree-walk(x)

if (x <> NIL)

then Inorder-Tree-Walk(left[x])

print key[x]

Inorder-Tree-walk(right[x])

A-B-C-D-E-F-G-H Listen To Part 8-10

Tree Insertion Do a binary search to find where it should be, then replace the termination NIL pointer with the new item.

file:///E|/LEC/LECTUR17/NODE8.HTM (6 of 13) [19/1/2003 1:34:46]

Lecture 8 - binary trees

Tree-insert(T,z)

y = NIL

x = root[T]

while

do y = x

if key[z] < key[x]

then x = left[x]

else x = right[x]

if y = NIL

then

else if key[z] < key[y]

then

else

y is maintained as the parent of x, since x eventually becomes NIL. The final test establishes whether the NIL was a left or right turn from y. Insertion takes time proportional to the height of the tree, O(h).

file:///E|/LEC/LECTUR17/NODE8.HTM (7 of 13) [19/1/2003 1:34:46]

Lecture 8 - binary trees

Listen To Part 8-12

Tree Deletion Deletion is somewhat more tricky than insertion, because the node to die may not be a leaf, and thus effect other nodes. Case (a), where the node is a leaf, is simple - just NIL out the parents child pointer. Case (b), where a node has one chld, the doomed node can just be cut out. Case (c), relabel the node as its successor (which has at most one child when z has two children!) and delete the successor! This implementation of deletion assumes parent pointers to make the code nicer, but if you had to save space they could be dispensed with by keeping the pointers on the search path stored in a stack.

Tree-Delete(T,z)

if (left[z] = NIL) or (right[z] = NIL)

then

else

Tree-Successor(z)

if

then

else

if

then

if p[y] = NIL

then

else if (y = left[p[y]]) file:///E|/LEC/LECTUR17/NODE8.HTM (8 of 13) [19/1/2003 1:34:46]

Lecture 8 - binary trees

then

else

if (y <> z)

then

/* If y has other fields, copy them, too. */

return y

Lines 1-3 determine which node y is physically removed. Lines 4-6 identify x as the non-nil decendant, if any. Lines 7-8 give x a new parent. Lines 9-10 modify the root node, if necessary Lines 11-13 reattach the subtree, if necessary. Lines 14-16 if the removed node is deleted, copy. Conclusion: deletion takes time proportional to the height of the tree. Listen To Part 8-13

Balanced Search Trees All six of our dictionary operations, when implemented with binary search trees, take O(h), where h is the height of the tree. The best height we could hope to get is

, if the tree was perfectly balanced, since

But if we get unlucky with our order of insertion or deletion, we could get linear height!

insert(a)

file:///E|/LEC/LECTUR17/NODE8.HTM (9 of 13) [19/1/2003 1:34:46]

Lecture 8 - binary trees

insert(b)

insert(c)

insert(d)

In fact, random search trees on average have

height, but we are worried about worst case height.

We can't easily use randomization - Why? Listen To Part 8-14

Perfectly Balanced Trees Perfectly balanced trees require a lot of work to maintain:

If we insert the key 1, we must move every single node in the tree to rebalance it, taking Therefore, when we talk about "balanced" trees, we mean trees whose height is delete, search, min/max, successor/predecessor) take time.

time.

, so all dictionary operations (insert,

Red-Black trees are binary search trees where each node is assigned a color, where the coloring scheme helps us maintain the . height as Listen To Part 8-15

file:///E|/LEC/LECTUR17/NODE8.HTM (10 of 13) [19/1/2003 1:34:46]

Lecture 8 - binary trees

Red-Black Tree Definition Red-black trees have the following properties: 1. 2. 3. 4.

Every node is colored either red or black. Every leaf (NIL pointer) is black. If a node is red then both its children are black. Every single path from a node to a decendant leaf contains the same number of black nodes.

Listen To Part 8-16

What does this mean? If the root of a red-black tree is black can we just color it red? No! For one of its children might be red.

If an arbitrary node is red can we color it black? No! Because now all nodes may not have the same black height.

What tree maximizes the number of nodes in a tree of black height h?

Listen To Part 8-17

What does a red-black tree with two real nodes look like? file:///E|/LEC/LECTUR17/NODE8.HTM (11 of 13) [19/1/2003 1:34:46]

Lecture 8 - binary trees

Not (1) - consecutive reds Not (2), (4) - Non-Uniform black height Listen To Part 8-18

Red-Black Tree Height Lemma: A red-black tree with n internal nodes has height at most

.

Proof: Our strategy; first we bound the number of nodes in any subtree, then we bound the height of any subtree. We claim that any subtree rooted at x has at least

- 1 internal nodes, where bh(x) is the black height of node x.

Proof, by induction:

Now assume it is true for all tree with black height < bh(x). If x is black, both subtrees have black height bh(x)-1. If x is red, the subtrees have black height bh(x). Therefore, the number of internal nodes in any subtree is

Listen To Part 8-19 Now, let h be the height of our red-black tree. At least half the nodes on any single path from root to leaf must be black if we ignore the root. Thus This implies that

and

, so ,so

. . height6pt width4pt

Therefore red-black trees have height at most twice optimal. We have a balanced search tree if we can maintain the red-black tree structure under insertion and deletion.

Next: Lecture 9 - catch Up: No Title Previous: Lecture 7 - elementary

file:///E|/LEC/LECTUR17/NODE8.HTM (12 of 13) [19/1/2003 1:34:46]

Lecture 8 - binary trees

Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR17/NODE8.HTM (13 of 13) [19/1/2003 1:34:46]

Lecture 9 - catch up

Next: Lecture 10 - tree Up: No Title Previous: Lecture 8 - binary

Lecture 9 - catch up Listen To Part 9-1 11-1 For each of the four types of linked lists in the following table, what is the asymptotic worst-case running time for each dynamic-set operation listed?

singly

singly doubly

doubly

unsorted sorted unsorted sorted Search(L, k)

O(N)

O(N) O(N)

O(N)-

Insert(L, x)

O(1)

O(N) O(1)

O(N)-

Delete(L, x)

O(N)*

O(N)* O(1)

O(1)

Successor(L, x)

O(N)

O(1)

O(N)

O(1)

Predecessor(L, x) O(N)

O(N) O(N)

O(1)

Minimum(L)

O(N)

O(1)

O(1)

Maximum(L)

O(N)

O(1)+ O(N)

● ● ●

O(N)

O(1)+

I need a pointer to the predecessor! (*) I need a pointer to the tail! (+) Only bottlenecks in otherwise perfect dictionary! (-)

Next: Lecture 10 - tree Up: No Title Previous: Lecture 8 - binary Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR17/NODE9.HTM [19/1/2003 1:34:47]

Lecture 10 - tree restructuring

Next: Lecture 11 - backtracking Up: No Title Previous: Lecture 9 - catch

Lecture 10 - tree restructuring Listen To Part 10-1 14.1-5 Describe a Red-Black tree with the largest and smallest ratio of red nodes.

To minimize the ratio of red-black nodes, make all black (possible for

To maximize the ratio of red nodes, interleave with red nodes as real leaves

Listen To Part 10-2

Rotations The basic restructuring step for binary search trees are left and right rotation:

file:///E|/LEC/LECTUR16/NODE10.HTM (1 of 9) [19/1/2003 1:34:50]

)

Lecture 10 - tree restructuring

1. Rotation is a local operation changing O(1) pointers. 2. An in-order search tree before a rotation stays an in-order search tree. 3. In a rotation, one subtree gets one level closer to the root and one subtree one level further from the root.

LEFT-ROTATE(T,x)

(* Set y*)

(* Turn y's left into x's right*)

if left[y]= NIL

then

(* Link x's parent to y *)

if p[x] = NIL

then

else if x= left[p[x]]

then

else

file:///E|/LEC/LECTUR16/NODE10.HTM (2 of 9) [19/1/2003 1:34:50]

Lecture 10 - tree restructuring

Note the in-order property is preserved. Listen To Part 10-3

Listen To Part 10-4 14.2-5 Show that any n-node tree can be transformed to any other using O(n) rotations (hint: convert to a right going chain).

I will start by showing weaker bounds - that first saw the problem.

and

First, observe that creating a right-going, for .

path from

rotations suffice - because that is how I proceeded when I

< and reversing the same construction gives a path from

to

Note that it will take at most n rotations to make the lowest valued key the root. Once it is root, all keys are to the right of it, so no more rotations need go through it to create a right-going chain. Repeating with the second lowest key, third, etc. rotations suffice. gives that Now that if we try to create a completely balanced tree instead. To get the n/2 key to the root takes at most n rotations. Now each subtree has half the nodes and we can recur...

file:///E|/LEC/LECTUR16/NODE10.HTM (3 of 9) [19/1/2003 1:34:50]

Lecture 10 - tree restructuring

Listen To Part 10-5 To get a linear algorithm, we must beware of trees like:

The correct answer is that n-1 rotations suffice to get to a rightmost chain. By picking the lowest node on the rightmost chain which has a left ancestor, we can add one node per rotation to the right most chain!

Initially, the rightmost chain contained at least 1 node, so after 1 rotations it contains all n. Slick! Listen To Part 10-6

Red-Black Insertion

file:///E|/LEC/LECTUR16/NODE10.HTM (4 of 9) [19/1/2003 1:34:50]

Lecture 10 - tree restructuring

Since red-black trees have a balanced tree!

height, if we can preserve all properties of such trees under insertion/deletion, we have

Suppose we just did a regular insertion. Under what conditions does it stay a red-black tree? Since every insertion take places at a leaf, we will change a black NIL pointer to a node with two black NIL pointers.

To preserve the black height of the tree, the new node must be red. If its new parent is black, we can stop, otherwise we must restructure! Listen To Part 10-7

How can we fix two reds in a row? It depends upon our uncle's color:

If our uncle is red, reversing our relatives' color either solves the problem or pushes it higher!

file:///E|/LEC/LECTUR16/NODE10.HTM (5 of 9) [19/1/2003 1:34:50]

Lecture 10 - tree restructuring

Note that after the recoloring: 1. The black height is unchanged. 2. The shape of the tree is unchanged. 3. We are done if our great-grandparent is black. If we get all the way to the root, recall we can always color a red-black tree's root black. We always will, so initially it was black, and so this process terminates. Listen To Part 10-8

The Case of the Black Uncle If our uncle was black, observe that all the nodes around us have to be black:

Solution - rotate right about B:

file:///E|/LEC/LECTUR16/NODE10.HTM (6 of 9) [19/1/2003 1:34:50]

Lecture 10 - tree restructuring

Since the root of the subtree is now black with the same black-height as before, we have restored the colors and can stop! Listen To Part 10-9 A double rotation can be required to set things up depending upon the left-right turn sequence, but the principle is the same. DOUBLE ROTATION ILLUSTRATION Listen To Part 10-10

Pseudocode and Figures Listen To Part 10-11

Deletion from Red-Black Trees Recall the three cases for deletion from a binary tree: Case (a) The node to be deleted was a leaf;

Case (b) The node to be deleted had one child;

file:///E|/LEC/LECTUR16/NODE10.HTM (7 of 9) [19/1/2003 1:34:51]

Lecture 10 - tree restructuring

Case (c) relabel to node as its successor and delete the successor.

Listen To Part 10-12

Deletion Color Cases Suppose the node we remove was red, do we still have a red-black tree? Yes! No two reds will be together, and the black height for each leaf stays the same. However, if the dead node y was black, we must give each of its decendants another black ancestor. If an appropriate node is red, we can simply color it black otherwise we must restructure. Case (a) black NIL becomes ``double black''; Case (b) red

becomes black and black

becomes ``double black'';

Case (c) red becomes black and black

becomes ``double black''.

Our goal will be to recolor and restructure the tree so as to get rid of the ``double black'' node. Listen To Part 10-13 In setting up any case analysis, we must be sure that: 1. All possible cases are covered. file:///E|/LEC/LECTUR16/NODE10.HTM (8 of 9) [19/1/2003 1:34:51]

Lecture 10 - tree restructuring

2. No case is covered twice. In the case analysis for red-black trees, the breakdown is: Case 1: The double black node x has a red brother. Case 2: x has a black brother and two black nephews. Case 3: x has a black brother, and its left nephew is red and its right nephew is black. Case 4: x has a black brother, and its right nephew is red (left nephew can be any color). Listen To Part 10-14

Conclusion Red-Black trees let us implement all dictionary operations in . Further, in no case are more than 3 rotations done to rebalance. Certain very advanced data structures have data stored at nodes which requires a lot of work to adjust after a rotation -- red-black trees ensure it won't happen often. Example: Each node represents the endpoint of a line, and is augmented with a list of segments in its subtree which it intersects. We will not study such complicated structures, however.

Next: Lecture 11 - backtracking Up: No Title Previous: Lecture 9 - catch Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR16/NODE10.HTM (9 of 9) [19/1/2003 1:34:51]

Lecture 11 - backtracking

Next: Lecture 12 - introduction Up: No Title Previous: Lecture 10 - tree

Lecture 11 - backtracking Parallel Bubblesort In order for me to give back your midterms, please form a line and sort yourselves in alphabetical order, from A to Z. There is traditionally a strong correlation between the midterm grades and the number of daily problems attempted: daily: 0, sum: 134, count: 3, avg: 44.67 daily: 1, sum: 0, count: 2, avg: 0.00 daily: 2, sum: 63, count: 1, avg: 63.00 daily: 3, sum: 194, count: 3, avg: 64.67 daily: 4, sum: 335, count: 5, avg: 67.00 daily: 5, sum: 489, count: 8, avg: 61.12 daily: 6, sum: 381, count: 6, avg: 63.50 daily: 7, sum: 432, count: 6, avg: 72.00 daily: 8, sum: 217, count: 3, avg: 72.33 daily: 9, sum: 293, count: 4, avg: 73.25 Listen To Part 11-2

Combinatorial Search We have seen how clever algorithms can reduce sorting from combinatorially explosive problems:

to

. However, the stakes are even higher for

The Traveling Salesman Problem Given a weighted graph, find the shortest cycle which visits each vertex once.

file:///E|/LEC/LECTUR16/NODE11.HTM (1 of 8) [19/1/2003 1:34:54]

Lecture 11 - backtracking

Applications include minimizing plotter movement, printed-circuit board wiring, transportation problems, etc. There is no known polynomial time algorithm (ie. for some fixed k) for this problem, so search-based algorithms are the only way to go if you need an optional solution. Listen To Part 11-3

But I want to use a Supercomputer Moving to a faster computer can only buy you a relatively small improvement: ●

●

●

●

Hardware clock rates on the fastest computers only improved by a factor of 6 from 1976 to 1989, from 12ns to 2ns. Moving to a machine with 100 processors can only give you a factor of 100 speedup, even if your job can be perfectly parallelized (but of course it can't). The fast Fourier algorithm (FFT) reduced computation from to . This is a speedup of 340 times on n=4096 and revolutionized the field of image processing. The fast multipole method for n-particle interaction reduced the computation from to O(n). This is a speedup of 4000 times on n=4096.

Listen To Part 11-4

Can Eight Pieces Cover a Chess Board? Consider the 8 main pieces in chess (king, queen, two rooks, two bishops, two knights). Can they be positioned on a chessboard so every square is threatened?

file:///E|/LEC/LECTUR16/NODE11.HTM (2 of 8) [19/1/2003 1:34:54]

Lecture 11 - backtracking

Only 63 square are threatened in this configuration. Since 1849, no one had been able to find an arrangement with bishops on different colors to cover all squares. Of course, this is not an important problem, but we will use it as an example of how to attack a combinatorial search problem. Listen To Part 11-5

How many positions to test? Picking a square for each piece gives us the bound:

Anything much larger than

is unreasonable to search on a modest computer in a modest amount of time.

However, we can exploit symmetry to save work. With reflections along horizontal, vertical, and diagonal axis, the queen can go in only 10 non-equivallent positions. Even better, we can restrict the white bishop to 16 spots and the queen to 16, while being certain that we get all distinct configurations.

Listen To Part 11-6 file:///E|/LEC/LECTUR16/NODE11.HTM (3 of 8) [19/1/2003 1:34:54]

Lecture 11 - backtracking

Backtracking Backtracking is a systematic way to go through all the possible configurations of a search space. In the general case, we assume our solution is a vector ordered set ,

where each element

is selected from a finite

and try to extend it by adding another element. After We build from a partial solution of length k extending it, we will test whether what we have so far is still possible as a partial solution. If it is still a candidate solution, great. If not, we delete

Compute

and try the next element from

:

, the set of candidate first elements of v.

k = 1

While k > 0 do

While

do

(*advance*)

= an element in

if (

) is solution, print!

k = k + 1

compute

k = k - 1

, the candidate kth elements given v.

(*backtrack*)

Listen To Part 11-7

Recursive Backtracking file:///E|/LEC/LECTUR16/NODE11.HTM (4 of 8) [19/1/2003 1:34:54]

Lecture 11 - backtracking

Recursion can be used for elegant and easy implementation of backtracking.

Backtrack(a, k)

if a is a solution, print(a)

else {

k = k +1

compute

while

do

= an element in

=

Backtrack(a, k)

}

Backtracking can easily be used to iterate through all subsets or permutations of a set. Backtracking ensures correctness by enumerating all possibilities. For backtracking to be efficient, we must prune the search space. Listen To Part 11-8

Constructing all Subsets How many subsets are there of an n-element set? To construct all

subsets, set up an array/vector of n cells, where the value of

file:///E|/LEC/LECTUR16/NODE11.HTM (5 of 8) [19/1/2003 1:34:54]

is either true or false, signifying

Lecture 11 - backtracking

whether the ith item is or is not in the subset. To use the notation of the general backtrack algorithm, What order will this generate the subsets of

, and v is a solution whenever

.

?

Listen To Part 11-9

Constructing all Permutations How many permutations are there of an n-element set? To construct all n! permutations, set up an array/vector of n cells, where the value of is an integer from 1 to n which has not appeared thus far in the vector, corresponding to the ith element of the permutation. To use the notation of the general backtrack algorithm,

, and v is a solution whenever

.

The n-Queens Problem The first use of pruning to deal with the combinatorial explosion was by the king who rewarded the fellow who discovered chess! In the eight Queens, we prune whenever one queen threatens another. Listen To Part 11-11

Covering the Chess Board In covering the chess board, we prune whenever we find there is a square which we cannot cover given the initial configuration! file:///E|/LEC/LECTUR16/NODE11.HTM (6 of 8) [19/1/2003 1:34:54]

Lecture 11 - backtracking

Specifically, each piece can threaten a certain maximum number of squares (queen 27, king 8, rook 14, etc.) Whenever the number of unthreated squares exceeds the sum of the maximum number of coverage remaining in unplaced squares, we can prune. As implemented by a graduate student project, this backtrack search eliminates are ordered by decreasing mobility.

of the search space, when the pieces

With precomputing the list of possible moves, this program could search 1,000 positions per second. But this is too slow!

Although we might further speed the program by an order of magnitude, we need to prune more nodes! By using a more clever algorithm, we eventually were able to prove no solution existed, in less than one day's worth of computing. You too can fight the combinatorial explosion! Listen To Part 11-12

The Backtracking Contest: Bandwidth The bandwidth problem takes as input a graph G, with n vertices and m edges (ie. pairs of vertices). The goal is to find a permutation of the vertices on the line which minimizes the maximum length of any edge.

The bandwidth problem has a variety of applications, including circuit layout, linear algebra, and optimizing memory usage in hypertext documents. The problem is NP-complete, meaning that it is exceedingly unlikely that you will be able to find an algorithm with polynomial worst-case running time. It remains NP-complete even for restricted classes of trees. Since the goal of the problem is to find a permutation, a backtracking program which iterates through all the n! possible permutations and computes the length of the longest edge for each gives an easy algorithm. But the goal of this assignment is to find as practically good an algorithm as possible. Listen To Part 12-4

Rules of the Game

file:///E|/LEC/LECTUR16/NODE11.HTM (7 of 8) [19/1/2003 1:34:54]

Lecture 11 - backtracking

1. Everyone must do this assignment separately. Just this once, you are not allowed to work with your partner. The idea is to think about the problem from scratch. 2. If you do not completely understand what the bandwidth of a graph is, you don't have the slightest chance of producing a working program. Don't be afraid to ask for a clarification or explanation!!!!! 3. There will be a variety of different data files of different sizes. Test on the smaller files first. Do not be afraid to create your own test files to help debug your program. 4. The data files are available via the course WWW page. 5. You will be graded on how fast and clever your program is, not on style. No credit will be given for incorrect programs. 6. The programs are to run on the whatever computer you have access to, although it must be vanilla enough that I can run the program on something I have access to. 7. You are to turn in a listing of your program, along with a brief description of your algorithm and any interesting optimizations, sample runs, and the time it takes on sample data files. Report the largest test file your program could handle in one minute or less of wall clock time. 8. The top five self-reported times / largest sizes will be collected and tested by me to determine the winner. Listen To Part 12-5

Producing Efficient Programs 1. Don't optimize prematurely: Worrying about recursion vs. iteration is counter-productive until you have worked out the best way to prune the tree. That is where the money is. 2. Choose your data structures for a reason: What operations will you be doing? Is case of insertion/deletion more crucial than fast retrieval? When in doubt, keep it simple, stupid (KISS). 3. Let the profiler determine where to do final tuning: Your program is probably spending time where you don't expect.

Next: Lecture 12 - introduction Up: No Title Previous: Lecture 10 - tree Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR16/NODE11.HTM (8 of 8) [19/1/2003 1:34:54]

Lecture 12 - introduction to dynamic programming

Next: Lecture 13 - dynamic Up: No Title Previous: Lecture 11 - backtracking

Lecture 12 - introduction to dynamic programming Listen To Part 12-1 15.1-5 Given an element x in an n-node order-statistic binary tree and a natural number i, how can the time. ith successor of x be determined in

This problem can be solved if our data structure supports two operations: ● ●

Rank(x) - what is the position of x in the total order of keys? Get(i) - what is the key in the ith position of the total order of keys?

What we are interested in is Get(Rank(x)+i). In an order statistic tree, each node x is labeled with the number of nodes contained in the subtree rooted in x.

Implementing both operations involves keeping track of how many nodes lie to the left of our path. Listen To Part 12-6

Optimization Problems In the algorithms we have studied so far, correctness tended to be easier than efficiency. In optimization problems, we are interested in finding a thing which maximizes or minimizes some function. file:///E|/LEC/LECTUR16/NODE12.HTM (1 of 9) [19/1/2003 1:34:58]

Lecture 12 - introduction to dynamic programming

In designing algorithms for optimization problem - we must prove that the algorithm in fact gives the best possible solution. Greedy algorithms, which makes the best local decision at each step, occasionally produce a global optimum - but you need a proof!

Dynamic Programming Dynamic Programming is a technique for computing recurrence relations efficiently by sorting partial results. Listen To Part 12-9

Computing Fibonacci Numbers

Implementing it as a recursive procedure is easy but slow! We keep calculating the same value over and over!

How slow is slow?

Thus

, and since our recursion tree has 0 and 1 as leaves, means we have

file:///E|/LEC/LECTUR16/NODE12.HTM (2 of 9) [19/1/2003 1:34:58]

calls!

Lecture 12 - introduction to dynamic programming

Listen To Part 12-10

What about Dynamic Programming? We can calculate

in linear time by storing small values:

For i=1 to n

Moral: we traded space for time. Dynamic programming is a technique for efficiently computing recurrences by storing partial results. Once you understand dynamic programming, it is usually easier to reinvent certain algorithms than try to look them up! Dynamic programming is best understood by looking at a bunch of different examples. I have found dynamic programming to be one of the most useful algorithmic techniques in practice: ● ● ●

Morphing in Computer Graphics Data Compression for High Density Bar Codes Utilizing Grammatical Constraints for Telephone Keypads

Listen To Part 12-11

Multiplying a Sequence of Matrices Suppose we want to multiply a long sequence of matrices

file:///E|/LEC/LECTUR16/NODE12.HTM (3 of 9) [19/1/2003 1:34:58]

.

Lecture 12 - introduction to dynamic programming

Multiplying an multiplications.

matrix by a

matrix (using the common algorithm) takes

We would like to avoid big intermediate matrices, and since matrix multiplication is associative, we can parenthesise however we want. Matrix multiplication is not communitive, so we cannot permute the order of the matrices without changing the result. Listen To Part 12-12

Example Consider

, where A is

, B is

, C is

, and D is

.

There are three possible parenthesizations:

The order makes a big difference in real computation. How do we find the best order? Let M(i,j) be the minimum number of multiplications necessary to compute The key observations are ● ●

The outermost parentheses partition the chain of matricies (i,j) at some k. The optimal parenthesization order has optimal ordering on either side of k.

Listen To Part 12-13 A recurrence for this is: file:///E|/LEC/LECTUR16/NODE12.HTM (4 of 9) [19/1/2003 1:34:58]

.

Lecture 12 - introduction to dynamic programming

If there are n matrices, there are n+1 dimensions. A direct recursive implementation of this will be exponential, since there is a lot of duplicated work as in the Fibonacci recurrence. Divide-and-conquer is seems efficient because there is no overlap, but ... There are only

substrings between 1 and n. Thus it requires only

space to store the optimal cost

for each of them. We can represent all the possibilities in a triangle matrix. We can also store the value of k in another triangle matrix to reconstruct to order of the optimal parenthesisation. The diagonal moves up to the right as the computation progresses. On each element of the kth diagonal |ji| = k. For the previous example: Listen To Part 13-3

Procedure MatrixOrder

for i=1 to n do M[i, j]=0

for diagonal=1 to n-1

for i=1 to n-diagonal do

j=i+diagonal

file:///E|/LEC/LECTUR16/NODE12.HTM (5 of 9) [19/1/2003 1:34:58]

Lecture 12 - introduction to dynamic programming

faster(i,j)=k

return [m(1, n)]

Procedure ShowOrder(i, j)

if (i=j) write (

)

else

k=factor(i, j)

write ``(''

ShowOrder(i, k)

write ``*''

ShowOrder (k+1, j)

write ``)''

Listen To Part 13-4

A dynamic programming solution has three components:

file:///E|/LEC/LECTUR16/NODE12.HTM (6 of 9) [19/1/2003 1:34:58]

Lecture 12 - introduction to dynamic programming

1. Formulate the answer as a recurrence relation or recursive algorithm. 2. Show that the number of different instances of your recurrence is bounded by a polynomial. 3. Specify an order of evaluation for the recurrence so you always have what you need. Listen To Part 13-5

Approximate String Matching A common task in text editing is string matching - finding all occurrences of a word in a text. Unfortunately, many words are mispelled. How can we search for the string closest to the pattern? Let p be a pattern string and T a text string over the same alphabet. A k-approximate match between P and T is a substring of T with at most k differences. Differences may be: 1. the corresponding characters may differ: KAT CAT 2. P is missing a character from T: CAAT CAT 3. T is missing a character from P: CT CAT Approximate Matching is important in genetics as well as spell checking. Listen To Part 13-6

A 3-Approximate Match A match with one of each of three edit operations is: P = unescessaraly T = unnecessarily Finding such a matching seems like a hard problem because we must figure out where you add blanks, but we can solve it with dynamic programming. D[i, j] = the minimum number of differences between

file:///E|/LEC/LECTUR16/NODE12.HTM (7 of 9) [19/1/2003 1:34:58]

and the segment of T ending at j.

Lecture 12 - introduction to dynamic programming

D[i, j] is the minimum of the three possible ways to extend smaller strings: 1. If then D[i-1, j-1] else D[i-1, j-1]+1 (corresponding characters do or do not match) 2. D[i-1, j]+1 (extra character in text - we do not advance the pattern pointer). 3. D[i, j-1]+1 (character in pattern which is not in text). Once you accept the recurrence it is easy. To fill each cell, we need only consider three other cells, not O(n) as in other examples. This means we need only store two rows of the table. The total time is O(mn). Listen To Part 13-10

Boundary conditions for string matching What should the value of D[0,i] be, corresponding to the cost of matching the first i characters of the text with none of the pattern? It depends. Are we doing string matching in the text or substring matching? ●

●

If you want to match all of the pattern against all of the text, this meant that would have to delete the first i characters of the pattern, so D[0,i] = i to pay the cost of the deletions. if we want to find the place in the text where the pattern occurs? We do not want to pay more of a cost if the pattern occurs far into the text than near the front, so it is important that starting cost be equal for all positions. In this case, D[0,i] = 0, since we pay no cost for deleting the first i characters of the text.

In both cases, D[i,0] = i, since we cannot excuse deleting the first i characters of the pattern without cost. Listen To Part 13-9

What do we return? If we want the cost of comparing all of the pattern against all of the text, such as comparing the spelling of two words, all we are interested in is D[n,m]. But what if we want the cheapest match between the pattern anywhere in the text? Assuming the initialization for substring matching, we seek the cheapest matching of the full pattern ending anywhere in the text. This means the cost equals .

file:///E|/LEC/LECTUR16/NODE12.HTM (8 of 9) [19/1/2003 1:34:58]

Lecture 12 - introduction to dynamic programming

This only gives the cost of the optimal matching. The actual alignment - what got matched, substituted, and deleted - can be reconstructed from the pattern/text and table without an auxiliary storage, once we have identified the cell with the lowest cost. Listen To Part 13-11

How much space do we need?

Do we need to keep all O(mn) cells, since if we evaluate the recurrence filling in the columns of the matrix from left to right, we will never need more than two columns of cells to do what we need. Thus O(m) space is sufficient to evaluate the recurrence without changing the time complexity at all. Unfortunately, because we won't have the full matrix we cannot reconstruct the alignment, as above. Saving space in dynamic programming is very important. Since memory on any computer is limited, O(nm) space is more of a bottleneck than O(nm) time. Fortunately, there is a clever divide-and-conquer algorithm which computes the actual alignment in O(nm) time and O(m) space.

Next: Lecture 13 - dynamic Up: No Title Previous: Lecture 11 - backtracking Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR16/NODE12.HTM (9 of 9) [19/1/2003 1:34:58]

Lecture 13 - dynamic programming applications

Next: Lecture 14 - data Up: No Title Previous: Lecture 12 - introduction

Lecture 13 - dynamic programming applications Listen To Part 13-1 16.3-5 Give an numbers.

algorithm to find the longest montonically increasing sequence in a sequence of n

Build an example first: (5, 2, 8, 7, 1, 6, 4) Ask yourself what would you like to know about the first n-1 elements to tell you the answer for the entire sequence? 1. The length of the longest sequence in . (seems obvious) 2. The length of the longest sequence will extend! (not as obvious - this is the idea!) Let

be the length of the longest sequence ending with the ith character: sequence 5 2 8 7 3 1 6 4 1 1 2 2 2 1 3 3

How do we compute i?

To find the longest sequence - we know it ends somewhere, so Length = Listen To Part 14-5

The Principle of Optimality file:///E|/LEC/LECTUR16/NODE13.HTM (1 of 7) [19/1/2003 1:35:01]

Lecture 13 - dynamic programming applications

To use dynamic programming, the problem must observe the principle of optimality, that whatever the initial state is, remaining decisions must be optimal with regard the state following from the first decision. Combinatorial problems may have this property but may use too much memory/time to be efficient.

Example: The Traveling Salesman Problem Let

be the cost of the optimal tour for i to 1 that goes thru each of the other cities once

Here there can be any subset of

instead of any subinterval - hence exponential.

Still, with other ideas (some type of pruning or best-first search) it can be effective for combinatorial search. Listen To Part 14-6

When can you use Dynamic Programming? Dynamic programming computes recurrences efficiently by storing partial results. Thus dynamic programming can only be efficient when there are not too many partial results to compute! There are n! permutations of an n-element set - we cannot use dynamic programming to store the best solution for each subpermutation. There are subsets of an n-element set - we cannot use dynamic programming to store the best solution for each. However, there are only n(n-1)/2 continguous substrings of a string, each described by a starting and ending point, so we can use it for string problems. There are only n(n-1)/2 possible subtrees of a binary search tree, each described by a maximum and minimum key, so we can use it for optimizing binary search trees. Dynamic programming works best on objects which are linearly ordered and cannot be rearranged characters in a string, matrices in a chain, points around the boundary of a polygon, the left-to-right order of leaves in a search tree.

file:///E|/LEC/LECTUR16/NODE13.HTM (2 of 7) [19/1/2003 1:35:01]

Lecture 13 - dynamic programming applications

Whenever your objects are ordered in a left-to-right way, you should smell dynamic programming! Listen To Part 14-7

Minimum Length Triangulation A triangulation of a polygon is a set of non-intersecting diagonals which partitions the polygon into diagonals.

The length of a triangulation is the sum of the diagonal lengths. We seek to find the minimum length triangulation. For a convex polygon, or part thereof:

Once we identify the correct connecting vertex, the polygon is partitioned into two smaller pieces, both of which must be triangulated optimally!

Evaluation proceeds as in the matrix multiplication example time if we evaluate the sections in order of increasing size.

file:///E|/LEC/LECTUR16/NODE13.HTM (3 of 7) [19/1/2003 1:35:01]

values of t, each of which takes O(j-i)

Lecture 13 - dynamic programming applications

What if there are points in the interior of the polygon? Listen To Part 14-8

Dynamic Programming and High Density Bar Codes Symbol Technology has developed a new design for bar codes, PDF-417 that has a capacity of several hundred bytes. What is the best way to encode text for this design?

They developed a complicated mode-switching data compression scheme.

file:///E|/LEC/LECTUR16/NODE13.HTM (4 of 7) [19/1/2003 1:35:01]

Lecture 13 - dynamic programming applications

Latch commands permanently put you in a different mode. Shift commands temporarily put you in a different mode. Listen To Part 14-9 Originally, Symbol used a greedy algorithm to encode a string, making local decisions only. We realized that for any prefix, you want an optimal encoding which might leave you in every possible mode.

the cost of encoding the ith character and ending up in node j). Our simple dynamic programming algorithm improved to capacity of PDF-417 by an average of

!

Listen To Part 14-10

Dynamic Programming and Morphing Morphing is the problem of creating a smooth series of intermediate images given a starting and ending image. file:///E|/LEC/LECTUR16/NODE13.HTM (5 of 7) [19/1/2003 1:35:01]

Lecture 13 - dynamic programming applications

The key problem is establishing a correspondence between features in the two images. You want to morph an eye to an eye, not an ear to an ear. We can do this matching on a line-by-line basis:

This should sound like string matching, but with a different set of operations: ●

●

●

Full run match: We may match run i on top to run j on bottom for a cost which is a function of the difference in the lengths of the two runs and their positions. Merging runs: We may match a string of consecutive runs on top to a run on bottom. The cost will be a function of the number of runs, their relative positions, and lengths. Listen To Part 14-11 Splitting runs: We may match a big run on top to a string of consecutive runs on the bottom. This is just the converse of the merge. Again, the cost will be a function of the number of runs, their relative positions, and lengths.

This algorithm was incorported into a morphing system, with the following results:

file:///E|/LEC/LECTUR16/NODE13.HTM (6 of 7) [19/1/2003 1:35:01]

Lecture 13 - dynamic programming applications

Next: Lecture 14 - data Up: No Title Previous: Lecture 12 - introduction Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR16/NODE13.HTM (7 of 7) [19/1/2003 1:35:01]

Lecture 14 - data structures for graphs

Next: Lecture 15 - DFS Up: No Title Previous: Lecture 13 - dynamic

Lecture 14 - data structures for graphs Listen To Part 14-1

Problem Solving Techniques Most important: make sure you understand exactly what the question is asking - if not, you have no hope of answer it!! Never be afraid to ask for another explanation of a problem until it is clear. Play around with the problem by constructing examples to get insight into it. Ask yourself questions. Does the first idea which comes into my head work? If not, why not? Am I using all information that I am given about the problem? Read Polya's book How to Solve it. Listen To Part 14-2 16-1: The Euclidean traveling-salesman problem is the problem of determining the shortest closed tour that connects a given set of n points in the plane. Bentley suggested simplifying the problem by restricting attention to bitonic tours, that is tours which start at the leftmost point, go strictly left to right to the rightmost point, and then go strictly right back to the starting point.

file:///E|/LEC/LECTUR16/NODE14.HTM (1 of 12) [19/1/2003 1:35:05]

Lecture 14 - data structures for graphs

Describe an algorithm for finding the optimal bitonic tour. You may assume that no two points have the same x-coordinate. (Hint: scan left to right, maintaining optimal possibilities for the two parts of the tour.) Make sure you understand what a bitonic tour is, or else it is hopeless. First of all, play with the problem. Why isn't it trivial? Listen To Part 14-3

Am I using all the information? Why will they let us assume that no two x-coordinates are the same? What does the hint mean? What happens if I scan from left to right? If we scan from left to right, we get an open tour which uses all points to the left of our scan line.

file:///E|/LEC/LECTUR16/NODE14.HTM (2 of 12) [19/1/2003 1:35:05]

Lecture 14 - data structures for graphs

In the optimal tour, the kth point is connected to exactly one point to the left of k. Once I decide which point that is, say x. I need the optimal partial tour where the two endpoints are x and k-1, because if it isn't optimal I could come up with a better one. Listen To Part 14-4 Hey, I have got a recurrence! And look, the two parameters which describe my optimal tour are the two endpoints. Let c[k,n] be the optimal cost partial tour where the two endpoints are k
c[0, 1]=d[0, 1]

c[n-1, n] takes O(n) to update, c[k, n] k
.

But this doesn't quite give the tour, but just an open tour. We simply must figure where the last edge to n file:///E|/LEC/LECTUR16/NODE14.HTM (3 of 12) [19/1/2003 1:35:05]

Lecture 14 - data structures for graphs

must go.

Listen To Part 15-1

Graphs A graph G consists of a set of vertices V together with a set E of vertex pairs or edges. Graphs are important because any binary relation is a graph, so graphs can be used to represent essentially any relationship. Example: A network of roads, with cities as vertices and roads between cities as edges.

Example: An electronic circuit, with junctions as vertices as components as edges.

To understand many problems, we must think of them in terms of graphs! Listen To Part 15-2

The Friendship Graph Consider a graph where the vertices are people, and there is an edge between two people if and only if they are friends.

file:///E|/LEC/LECTUR16/NODE14.HTM (4 of 12) [19/1/2003 1:35:05]

Lecture 14 - data structures for graphs

This graph is well-defined on any set of people: SUNY SB, New York, or the world. What questions might we ask about the friendship graph? ●

If I am your friend, does that mean you are my friend?

●

A graph is undirected if (x,y) implies (y,x). Otherwise the graph is directed. The ``heard-of'' graph is directed since countless famous people have never heard of me! The ``had-sex-with'' graph is presumably undirected, since it requires a partner. Am I my own friend? An edge of the form (x,x) is said to be a loop. If x is y's friend several times over, that could be modeled using multiedges, multiple edges between the same pair of vertices. A graph is said to be simple if it contains no loops and multiple edges.

●

Listen To Part 15-3 Am I linked by some chain of friends to the President? A path is a sequence of edges connecting two vertices. Since Mel Brooks is my father's-sister'shusband's cousin, there is a path between me and him!

●

How close is my link to the President?

●

If I were trying to impress you with how tight I am with Mel Brooks, I would be much better off saying that Uncle Lenny knows him than to go into the details of how connected I am to Uncle Lenny. Thus we are often interested in the shortest path between two nodes. Is there a path of friends between any two people? A graph is connected if there is a path between any two vertices. A directed graph is strongly

file:///E|/LEC/LECTUR16/NODE14.HTM (5 of 12) [19/1/2003 1:35:05]

Lecture 14 - data structures for graphs

●

connected if there is a directed path between any two vertices. Who has the most friends? The degree of a vertex is the number of edges adjacent to it. Listen To Part 15-4

●

What is the largest clique?

●

A social clique is a group of mutual friends who all hang around together. A graph theoretic clique is a complete subgraph, where each vertex pair has an edge between them. Cliques are the densest possible subgraphs. Within the friendship graph, we would expect that large cliques correspond to workplaces, neighborhoods, religious organizations, schools, and the like. How long will it take for my gossip to get back to me? A cycle is a path where the last vertex is adjacent to the first. A cycle in which no vertex repeats (such as 1-2-3-1 verus 1-2-3-2-1) is said to be simple. The shortest cycle in the graph defines its girth, while a simple cycle which passes through each vertex is said to be a Hamiltonian cycle.

Listen To Part 15-5

Data Structures for Graphs There are two main data structures used to represent graphs.

Adjacency Matrices An adjacency matrix is an

It takes

matrix, where M[i,j] = 0 iff there is no edge from vertex i to vertex j

time to test if (i,j) is in a graph represented by an adjacency matrix.

Can we save space if (1) the graph is undirected? (2) if the graph is sparse?

file:///E|/LEC/LECTUR16/NODE14.HTM (6 of 12) [19/1/2003 1:35:05]

Lecture 14 - data structures for graphs

Listen To Part 15-6

Adjacency Lists An adjacency list consists of a edges incident on vertex i.

array of pointers, where the ith element points to a linked list of the

, where

To test if edge (i,j) is in the graph, we search the ith list for j, which takes the ith vertex.

is the degree of

Note that can be much less than n when the graph is sparse. If necessary, the two copies of each edge can be linked by a pointer to facilitate deletions. Listen To Part 15-7

Tradeoffs Between Adjacency Lists and Adjacency Matrices Comparison

Winner

Faster to test if (x, y) exists?

matrices

Faster to find vertex degree?

lists

Less memory on small graphs? Less memory on big graphs? Edge insertion or deletion? Faster to traverse the graph? Better for most problems?

lists (m+n) vs. matrices (small win) matrices O(1) lists m+n vs. lists

Both representations are very useful and have different properties, although adjacency lists are probably better for most problems. Listen To Part 16-2

file:///E|/LEC/LECTUR16/NODE14.HTM (7 of 12) [19/1/2003 1:35:05]

Lecture 14 - data structures for graphs

Traversing a Graph One of the most fundamental graph problems is to traverse every edge and vertex in a graph. Applications include: ● ● ●

Printing out the contents of each edge and vertex. Counting the number of edges. Identifying connected components of a graph.

For efficiency, we must make sure we visit each edge at most twice. For correctness, we must do the traversal in a systematic way so that we don't miss anything. Since a maze is just a graph, such an algorithm must be powerful enough to enable us to get out of an arbitrary maze. Listen To Part 16-3

Marking Vertices The idea in graph traversal is that we must mark each vertex when we first visit it, and keep track of what have not yet completely explored. For each vertex, we can maintain two flags: ● ●

discovered - have we ever encountered this vertex before? completely-explored - have we finished exploring this vertex yet?

We must also maintain a structure containing all the vertices we have discovered but not yet completely explored. Initially, only a single start vertex is considered to be discovered. To completely explore a vertex, we look at each edge going out of it. For each edge which goes to an undiscovered vertex, we mark it discovered and add it to the list of work to do. Note that regardless of what order we fetch the next vertex to explore, each edge is considered exactly twice, when each of its endpoints are explored. Listen To Part 16-4

file:///E|/LEC/LECTUR16/NODE14.HTM (8 of 12) [19/1/2003 1:35:05]

Lecture 14 - data structures for graphs

Correctness of Graph Traversal Every edge and vertex in the connected component is eventually visited. Suppose not, ie. there exists a vertex which was unvisited whose neighbor was visited. This neighbor will eventually be explored so we would visit it:

Listen To Part 16-5

Traversal Orders The order we explore the vertices depends upon what kind of data structure is used: ●

●

Queue - by storing the vertices in a first-in, first out (FIFO) queue, we explore the oldest unexplored vertices first. Thus our explorations radiate out slowly from the starting vertex, defining a so-called breadth-first search. Stack - by storing the vertices in a last-in, first-out (LIFO) stack, we explore the vertices by lurching along a path, constantly visiting a new neighbor if one is available, and backing up only if we are surrounded by previously discovered vertices. Thus our explorations quickly wander away from our starting point, defining a so-called depth-first search.

The three possible colors of each node reflect if it is unvisited (white), visited but unexplored (grey) or completely explored (black). Listen To Part 16-6

Breadth-First Search

BFS(G,s)

for each vertex

do

file:///E|/LEC/LECTUR16/NODE14.HTM (9 of 12) [19/1/2003 1:35:05]

Lecture 14 - data structures for graphs

color[u] = white

, ie. the distance from

p[u] = NIL, ie. the parent in the BFS tree

color[u] = grey

d[] = 0

p[] = NIL

while

do

u = head[Q]

for each

do

if color[v] = white then

color[v] = gray

d[v] = d[u] + 1

p[v] = u

file:///E|/LEC/LECTUR16/NODE14.HTM (10 of 12) [19/1/2003 1:35:05]

Lecture 14 - data structures for graphs

enqueue[Q,v]

dequeue[Q]

color[u] = black Listen To Part 16-8

Depth-First Search DFS has a neat recursive implementation which eliminates the need to explicitly use a stack. Discovery and final times are sometimes a convenience to maintain.

DFS(G)

for each vertex

do

color[u] = white

parent[u] = nil

time = 0

for each vertex

do

if color[u] = white then DFS-VISIT[u]

Initialize each vertex in the main routine, then do a search from each connected component. BFS must also start from a vertex in each component to completely visit the graph. file:///E|/LEC/LECTUR16/NODE14.HTM (11 of 12) [19/1/2003 1:35:05]

Lecture 14 - data structures for graphs

DFS-VISIT[u]

color[u] = grey (*u had been white/undiscovered*)

discover[u] = time

time = time+1

for each

do

if color[v] = white then

parent[v] = u

DFS-VISIT(v)

color[u] = black (*now finished with u*)

finish[u] = time

time = time+1

Next: Lecture 15 - DFS Up: No Title Previous: Lecture 13 - dynamic Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR16/NODE14.HTM (12 of 12) [19/1/2003 1:35:05]

Lecture 15 - DFS and BFS

Next: Lecture 16 - applications Up: No Title Previous: Lecture 14 - data

Lecture 15 - DFS and BFS Listen To Part 15-8 such that 23.1-5 - The square of a directed graph G=(V,E) is the graph , both and ; ie. there is a path of exactly two edges. some

iff for

Give efficient algorithms for both adjacency lists and matricies.

Given an adjacency matrix, we can check in constant time whether a given edge exists. To discover , for each possible intermediate vertex v we can check whether (u,v) whether there is an edge and (v,w) exist in O(1). Since there are at most n intermediate vertices to check, and time.

pairs of vertices to ask about, this takes

With adjacency lists, we have a list of all the edges in the graph. For a given edge (u,v), we can run through all the edges from v in O(n) time, and fill the results into an adjacency matrix of , which is initially empty. It takes O(mn) to construct the edges, and to initialize and read the adjacency matrix, a total of unless the graph is disconnected, this is usually simplified to O(mn), and is O((n+m)n). Since faster than the previous algorithm on sparse graphs. Why is it called the square of a graph? Because the square of the adjacency matrix is the adjacency matrix of the square! This provides a theoretically faster algorithm. Listen To Part 16-10

BFS Trees If BFS is performed on a connected, undirected graph, a tree is defined by the edges involved with the discovery of new nodes:

file:///E|/LEC/LECTUR16/NODE15.HTM (1 of 8) [19/1/2003 1:35:08]

Lecture 15 - DFS and BFS

This tree defines a shortest path from the root to every other node in the tree. The proof is by induction on the length of the shortest path from the root: ●

●

Length = 1 First step of BFS explores all neighbors of the root. In an unweighted graph one edge must be the shortest path to any node. Length = s Assume the BFS tree has the shortest paths up to length s-1. Any node at a distance of will first be discovered by expanding a distance s-1 node.

Listen To Part 16-11

The key idea about DFS A depth-first search of a graph organizes the edges of the graph in a precise way. In a DFS of an undirected graph, we assign a direction to each edge, from the vertex which discover it:

In a DFS of a directed graph, every edge is either a tree edge or a black edge.

file:///E|/LEC/LECTUR16/NODE15.HTM (2 of 8) [19/1/2003 1:35:08]

Lecture 15 - DFS and BFS

In a DFS of a directed graph, no cross edge goes to a higher numbered or rightward vertex. Thus, no edge from 4 to 5 is possible:

Listen To Part 16-12

Edge Classification for DFS What about the other edges in the graph? Where can they go on a search? Every edge is either:

On any particular DFS or BFS of a directed or undirected graph, each edge gets classified as one of the above. Listen To Part 17-3

DFS Trees The reason DFS is so important is that it defines a very nice ordering to the edges of the graph.

file:///E|/LEC/LECTUR16/NODE15.HTM (3 of 8) [19/1/2003 1:35:08]

Lecture 15 - DFS and BFS

In a DFS of an undirected graph, every edge is either a tree edge or a back edge. Why? Suppose we have a forward edge. We would have encountered (4,1) when expanding 4, so this is a back edge.

Suppose we have a cross-edge

Paths in search trees Where is the shortest path in a DFS?

file:///E|/LEC/LECTUR16/NODE15.HTM (4 of 8) [19/1/2003 1:35:08]

Lecture 15 - DFS and BFS

It could use multiple back and tree edges, where BFS only uses tree edges. DFS gives a better approximation of the longest path than BFS.

Listen To Part 17-4

Topological Sorting A directed, acyclic graph is a directed graph with no directed cycles. file:///E|/LEC/LECTUR16/NODE15.HTM (5 of 8) [19/1/2003 1:35:08]

Lecture 15 - DFS and BFS

A topological sort of a graph is an ordering on the vertices so that all edges go from left to right. Only a DAG can have a topological sort.

Any DAG has (at least one) topological sort. Listen To Part 17-5

Applications of Topological Sorting Topological sorting is often useful in scheduling jobs in their proper sequence. In general, we can use it to order things given constraints, such as a set of left-right constraints on the positions of objects. Example: Dressing schedule from CLR. Example: Identifying errors in DNA fragment assembly. Certain fragments are constrained to be to the left or right of other fragments, unless there are errors.

Solution - build a DAG representing all the left-right constraints. Any topological sort of this DAG is a consistant ordering. If there are cycles, there must be errors. A DFS can test if a graph is a DAG (it is iff there are no back edges - forward edges are allowed for DFS on directed graph). Listen To Part 17-6

file:///E|/LEC/LECTUR16/NODE15.HTM (6 of 8) [19/1/2003 1:35:08]

Lecture 15 - DFS and BFS

Algorithm Theorem: Arranging vertices in decreasing order of DFS finishing time gives a topological sort of a DAG. Proof: Consider any directed edge u,v, when we encounter it during the exploration of vertex u: ● ● ●

If v is white - we then start a DFS of v before we continue with u. If v is grey - then u, v is a back edge, which cannot happen in a DAG. If v is black - we have already finished with v, so f[v]
Thus we can do topological sorting in O(n+m) time. Listen To Part 17-8

Articulation Vertices Suppose you are a terrorist, seeking to disrupt the telephone network. Which station do you blow up?

An articulation vertex is a vertex of a connected graph whose deletion disconnects the graph. Clearly connectivity is an important concern in the design of any network. Articulation vertices can be found in O(n(m+n)) - just delete each vertex to do a DFS on the remaining graph to see if it is connected. Listen To Part 17-9

A Faster O(n+m) DFS Algorithm Theorem: In a DFS tree, a vertex v (other than the root) is an articulation vertex iff v is not a leaf and some subtree of v has no back edge incident until a proper ancestor of v.

file:///E|/LEC/LECTUR16/NODE15.HTM (7 of 8) [19/1/2003 1:35:08]

Lecture 15 - DFS and BFS

Proof: (1) v is an articulation vertex

v cannot be a leaf.

Why? Deleting v must seperate a pair of vertices x and y. Because of the other tree edges, this cannot happen unless y is a decendant of v. Listen To Part 17-10

v separating x,y implies there is no back edge in the subtree of y to a proper ancestor of v. (2) Conditions v is a non-root articulation vertex. v separates any ancestor of v from any decendant in the appropriate subtree. Actually implementing this test in O(n+m) is tricky - but believable once you accept this theorem.

Next: Lecture 16 - applications Up: No Title Previous: Lecture 14 - data Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR16/NODE15.HTM (8 of 8) [19/1/2003 1:35:08]

Lecture 16 - applications of DFS and BFS

Next: Lecture 17 - minimum Up: No Title Previous: Lecture 15 - DFS

Lecture 16 - applications of DFS and BFS Listen To Part 16-1 23.2-6 Give an efficient algorithm to test if a graph is bipartite.

Bipartite means the vertices can be colored red or black such that no edge links vertices of the same color.

Suppose we color a vertex red - what color must its neighbors be? black! We can augment either BFS or DFS when we first discover a new vertex, color it opposited its parents, and for each other edge, check it doesn't link two vertices of the same color. The first vertex in any connected component can be red or black! Bipartite graphs arise in many situations, and special algorithms are often available for them. What is the interpretation of a bipartite ``had-sex-with'' graph? How would you break people into two groups such that no group contains a pair of people who hate each other? Listen To Part 17-1 23.4-3 Give an O(n) algorithm to test whether an undirected graph contains a cycle. file:///E|/LEC/LECTUR16/NODE16.HTM (1 of 5) [19/1/2003 1:35:10]

Lecture 16 - applications of DFS and BFS

If you do a DFS, you have a cycle iff you have a back edge. This gives an O(n+m) algorithm. But where does the m go? If the graph contains more than n-1 edges, it must contain a cycle! Thus we never need look at more than n edges if we are given an adjacency list representation! Listen To Part 17-7 23.4-5 Show that you can topologically sort in O(n+m) by repeatedly deleting vertices of degree 0.

The correctness of this algorithm follows since in a DAG there must always be a vertex of indegree 0, and such a vertex can be first in topological sort. Suppose each vertex is initialized with its indegree (do DFS on G to get this). Deleting a vertex takes O(degree v). Reduce the indegree of each efficient vertex and keep a list of degree-0 vertices to delete next. Time: Listen To Part 17-12

Strongly Connected Components A directed graph is strongly connected iff there is a directed path between any two vertices. The strongly connected components of a graph is a partition of the vertices into subsets (maximal) such that each subset is strongly connected.

Observe that no vertex can be in two maximal components, so it is a partition.

file:///E|/LEC/LECTUR16/NODE16.HTM (2 of 5) [19/1/2003 1:35:10]

Lecture 16 - applications of DFS and BFS

There is an amazingly elegant, linear time algorithm to find the strongly connected components of a directed graph, using DFS. Listen To Part 17-13 ● ● ● ●

Call DFS( ) to compute finishing times for each vertex. Compute the transpose graph (reverse all edges in G) Call DFS( ), but order the vertices in decreasing order of finish time. The vertices of each DFS tree in the forest of DFS( ) is a strongly connected component.

This algorithm takes O(n+m), but why does it compute strongly connected components? Lemma: If two vertices are in the same strong component, no path between them ever leaves the component.

Lemma: In any DFS forest, all vertices in the same strongly connected component are in the same tree. Proof: Consider the first vertex v in the component to be discovered. Everything in the component is reachable from it, so we will traverse it before finishing with v. Listen To Part 17-14

What does DFS(

, v) Do?

It tells you what vertices have directed paths to v, while DFS( ,v) tells what vertices have directed paths , v) also have a path from u? from v. But why must any vertex in the search tree of DFS(

file:///E|/LEC/LECTUR16/NODE16.HTM (3 of 5) [19/1/2003 1:35:10]

Lecture 16 - applications of DFS and BFS

Because there is no edge from any previous DFS tree into the last tree!! Because we ordered the vertices by decreasing order of finish time, we can peel off the strongly connected components from right to left just be doing a DFS( ). Listen To Part 17-16

Example of Strong Components Algorithm

9, 10, 11, 12 can reach 9, oldest remaining finished is 5. 5, 6, 8 can reach 5, oldest remaining is 7. 7 can reach 7, oldest remaining is 1. 1, 2, 3 can reach 1, oldest remaining is 4. 4 can reach 4.

file:///E|/LEC/LECTUR16/NODE16.HTM (4 of 5) [19/1/2003 1:35:10]

Lecture 16 - applications of DFS and BFS

Next: Lecture 17 - minimum Up: No Title Previous: Lecture 15 - DFS Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR16/NODE16.HTM (5 of 5) [19/1/2003 1:35:10]

Lecture 17 - minimum spanning trees

Next: Lecture 18 - shortest Up: No Title Previous: Lecture 16 - applications

Lecture 17 - minimum spanning trees Listen To Part 19-4 24.2-6 Describe an efficent algorithm that, given an undirected graph G, determines a spanning tree G whose largest edge weight is minimum over all spanning trees of G.

First, make sure you understand the question

``Hey, doesn't Kruskal's algorithm do something like this.'' Certainly! Since Krushal's algorithm considers the edges in order of increasing weight, and stops the moment these edges form a connected graph, the tree it gives must minimize the edge weight. ``Hey, but then why doesn't Prim's algorithm also work?'' It gives the same thing as Kruskal's algorithm, so it must be true that any minimum spanning tree minimizes the maximum edge weight! Proof: Give me a MST and consider the largest edge weight, Listen To Part 19-5

file:///E|/LEC/LECTUR16/NODE17.HTM (1 of 11) [19/1/2003 1:35:14]

Lecture 17 - minimum spanning trees

Deleting it disconnects the MST. If there was a lower edge connects the two subtrees, I didn't have a MST! Listen To Part 18-2

Minimum Spanning Trees A tree is a connected graph with no cycles. A spanning tree is a subgraph of G which has the same set of vertices of G and is a tree. A minimum spanning tree of a weighted graph G is the spanning tree of G whose edges sum to minimum weight. There can be more than one minimum spanning tree in a graph edges.

consider a graph with identical weight

The minimum spanning tree problem has a long history - the first algorithm dates back at least to 1926!. Minimum spanning tree is always taught in algorithm courses since (1) it arises in many applications, (2) it is an important example where greedy algorithms always give the optimal answer, and (3) Clever data structures are necessary to make it work. In greedy algorithms, we make the decision of what next to do by selecting the best local option from all available choices - without regard to the global structure. Listen To Part 18-3

Applications of Minimum Spanning Trees Minimum spanning trees are useful in constructing networks, by describing the way to connect a set of sites using the smallest total amount of wire. Much of the work on minimum spanning (and related Steiner) trees has been conducted by the phone company. Minimum spanning trees provide a reasonable way for clustering points in space into natural groups. When the cities are points in the Euclidean plane, the minimum spanning tree provides a good heuristic for traveling salesman problems. The optimum traveling salesman tour is at most twice the length of the minimum spanning tree.

file:///E|/LEC/LECTUR16/NODE17.HTM (2 of 11) [19/1/2003 1:35:14]

Lecture 17 - minimum spanning trees

Listen To Part 18-4

Prim's Algorithm If G is connected, every vertex will appear in the minimum spanning tree. If not, we can talk about a minimum spanning forest. Prim's algorithm starts from one vertex and grows the rest of the tree an edge at a time. As a greedy algorithm, which edge should we pick? The cheapest edge with which can grow the tree by one vertex without creating a cycle. During execution we will label each vertex as either in the tree, fringe - meaning there exists an edge from a tree vertex, or unseen - meaning the vertex is more than one edge away.

Select an arbitrary vertex to start.

While (there are fringe vertices)

select minimum weight edge between tree and fringe

add the selected edge and vertex to the tree

Clearly this creates a spanning tree, since no cycle can be introduced via edges between tree and fringe vertices, but is it minimum? file:///E|/LEC/LECTUR16/NODE17.HTM (3 of 11) [19/1/2003 1:35:14]

Lecture 17 - minimum spanning trees

Listen To Part 18-5

Why is Prim's algorithm correct? Don't be scared by the proof - the reason is really quite basic: Theorem: Let G be a connected, weighted graph and let be a subset of the edges in a MST . Let V' be the vertices incident with edges in E'. If (x,y) is an edge of minimum weight such that and y is not in V', then is a subset of a minimum spanning tree. Proof: If the edge is in T, this is trivial. Suppose (x,y) is not in T Then there must be a path in T from x to y since T is connected. If (v,w) is the first edge on this path with one edge in V', if we delete it and replace it with (x, y) we get a spanning tree. This tree must have smaller weight than T, since W(v,w)>W(x,y). Thus T could not have been the MST.

Prim's Algorithm is correct! Thus we cannot go wrong with the greedy strategy the way we could with the traveling salesman problem. Listen To Part 18-6

But how fast is Prim's? That depends on what data structures are used. In the simplest implementation, we can simply mark each vertex as tree and non-tree and search always from scratch: file:///E|/LEC/LECTUR16/NODE17.HTM (4 of 11) [19/1/2003 1:35:14]

Lecture 17 - minimum spanning trees

Select an arbitrary vertex to start.

While (there are non-tree vertices)

select minimum weight edge between tree and fringe

add the selected edge and vertex to the tree

This can be done in O(n m) time, by doing a DFS or BFS to loop through all edges, with a constant time test per edge, and a total of n iterations. Can we do faster? If so, we need to be able to identify fringe vertices and the minimum cost edge associated with it, fast. We will augment an adjacency list with fields maintaining fringe information. Vertex: fringelink pointer to next vertex in fringe list. fringe weight cheapest edge linking v to l. parent other vertex with v having fringeweight. status intree, fringe, unseen. adjacency list the list of edges. Listen To Part 18-8 Finding the minimum weight fringe-edge takes O(n) time - just bump through fringe list. After adding a vertex to the tree, running through its adjacency list to update the cost of adding fringe vertices (there may be a cheaper way through the new vertex) can be done in O(n) time. Total time is

.

Listen To Part 18-9

Kruskal's Algorithm file:///E|/LEC/LECTUR16/NODE17.HTM (5 of 11) [19/1/2003 1:35:14]

Lecture 17 - minimum spanning trees

Since an easy lower bound argument shows that every edge must be looked at to find the minimum spanning tree, and the number of edges , Prim's algorithm is optimal in the worst case. Is that all she wrote? The complexity of Prim's algorithm is independent of the number of edges. Can we do better with sparse graphs? Yes! Kruskal's algorithm is also greedy. It repeatedly adds the smallest edge to the spanning tree that does not create a cycle. Obviously, this gives a spanning tree, but is it minimal? Listen To Part 18-10

Why is Kruskal's algorithm correct? Theorem: Let G be a weighted graph and let edge in E-E' which does not create a cycle,

. If E' is contained in a MST T and e is the smallest .

Proof: As before, suppose e is not in T. Adding e to T makes a cycle. Deleting another edge from this cycle leaves a connected graph, and if it is one from E-E' the cost of this tree goes down. Since such an edge exists, T could not be a MST.

Listen To Part 18-11

How fast is Kruskal's algorithm? What is the simplest implementation? ● ●

Sort the m edges in time. For each edge in order, test whether it creates a cycle the forest we have thus far built - if so discard, else add to forest. With a BFS/DFS, this can be done in O(n) time (since the tree has at most n edges).

file:///E|/LEC/LECTUR16/NODE17.HTM (6 of 11) [19/1/2003 1:35:14]

Lecture 17 - minimum spanning trees

The total time is O(mn), but can we do better? Kruskal's algorithm builds up connected components. Any edge where both vertices are in the same connected component create a cycle. Thus if we can maintain which vertices are in which component fast, we do not have test for cycles!

Put the edges in a heap

count=0

while (count < n-1) do

get next edge (v,w)

if (component (v)

component(w))

add to T

component (v)=component(w)

If we can test components in Question: Is

better than

, we can find the MST in

!

?

Listen To Part 19-6

Union-Find Programs Our analysis that Kruskal's MST algorithm is two vertices in the same connected component.

requires a fast way to test whether an edge links

Thus we need a data structure for maintaining sets which can test if two elements are in the same and file:///E|/LEC/LECTUR16/NODE17.HTM (7 of 11) [19/1/2003 1:35:14]

Lecture 17 - minimum spanning trees

merge two sets together. These can be implemented by UNION and FIND operations:

Is

t= Find

u= Find

Return (Is t=u?)

Make

Union(t, u)

Find returns the name of the set and Union sets the members of t to have the same name as u. We are interested in minimizing the time it takes to execute any sequence of unions and finds. A simple implementation is to represent each set as a tree, with pointers from a node to its parent. Each element is contained in a node, and the name of the set is the key at the root: Listen To Part 19-7

file:///E|/LEC/LECTUR16/NODE17.HTM (8 of 11) [19/1/2003 1:35:14]

Lecture 17 - minimum spanning trees

In the worst case, these structures can be very unbalanced:

For i = 1 to n/2 do

UNION(i,i+1)

For i = 1 to n/2 do

FIND(1)

We want the limit the height of our trees which are effected by UNIONs. When we union, we can make the tree with fewer nodes the child. Since the number of nodes is related to the height, the height of the final tree will increase only if both subtrees are of equal height! Lemma: If Union(t,v) attaches the root of v as a subtree of t iff the number of nodes in t is greater than or equal to the number in v, after any sequence of unions, any tree with h/4 nodes has height at most . Listen To Part 19-8 Proof: By induction on the number of nodes k, k=1 has height 0.

file:///E|/LEC/LECTUR16/NODE17.HTM (9 of 11) [19/1/2003 1:35:14]

Lecture 17 - minimum spanning trees

Assume true to k-1 nodes. Let

If

then

If

, then

be the height of the tree

.

Listen To Part 19-9

Can we do better? We can do unions and finds in

, good enough for Kruskal's algorithm. But can we do better?

The ideal Union-Find tree has depth 1:

On a find, if we are going down a path anyway, why not change the pointers to point to the root?

This path compression will let us do better than

file:///E|/LEC/LECTUR16/NODE17.HTM (10 of 11) [19/1/2003 1:35:14]

for n union-finds.

Lecture 17 - minimum spanning trees

O(n)? Not quite ... Difficult analysis shows that it takes time, where Ackerman function and number of atoms in the universe)=5.

Next: Lecture 18 - shortest Up: No Title Previous: Lecture 16 - applications Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR16/NODE17.HTM (11 of 11) [19/1/2003 1:35:14]

is the inverse

Lecture 18 - shortest path algorthms

Next: Lecture 19 - satisfiability Up: No Title Previous: Lecture 17 - minimum

Lecture 18 - shortest path algorthms Listen To Part 20-7 25.1-1 Give two more shortest path trees for the following graph:

Run through Dijkstra's algorithm, and see where there are ties which can be arbitrarily selected. There are two choices for how to get to the third vertex x, both of which cost 5. There are two choices for how to get to vertex v, both of which cost 9. Listen To Part 19-1

Lessons from the Backtracking contest ●

●

●

●

As predicted, the speed difference between the fastest programs and average program dwarfed the difference between a supercomputer and a microcomputer. Algorithms have a bigger impact on performance than hardware! Different algorithms perform differently on different data. Thus even hard problems may be tractable on the kind of data you might be interested in. None of the programs could efficiently handle all instances for . We will find out why after the midterm, when we discuss NP-completeness. Many of the fastest programs were very short and simple (KISS). My bet is that many of the enhancements students built into them actually showed them down! This is where profiling can come in handy.

file:///E|/LEC/LECTUR16/NODE18.HTM (1 of 10) [19/1/2003 1:35:18]

Lecture 18 - shortest path algorthms ●

The fast programs were often recursive.

Listen To Part 19-3

Winning Optimizations ●

● ●

●

●

Finding a good initial solution via randomization or heuristic improvement helped by establishing a good upper bound, to constrict search. Using half the largest vertex degree as a lower bound similarly constricted search. Pruning a partial permutation the instant an edge was the target made the difference in going from (say) 8 to 18. Positioning the partial permutation vertices separated by b instead of 1 meant significantly earlier cutoffs, since any edge does the job. Mirror symmetry can only save a factor of 2, but perhaps more could follow from partitioning the vertices into equivalence classes by the same neighborhood.

Listen To Part 19-10

Shortest Paths Finding the shortest path between two nodes in a graph arises in many different applications: ● ●

●

Transportation problems - finding the cheapest way to travel between two locations. Motion planning - what is the most natural way for a cartoon character to move about a simulated environment. Communications problems - how look will it take for a message to get between two places? Which two locations are furthest apart, ie. what is the diameter of the network.

Listen To Part 20-1

Shortest Paths and Sentence Disambiguation In our work on reconstructing text typed on an (overloaded) telephone keypad, we had to select which of many possible interpretations was most likely.

file:///E|/LEC/LECTUR16/NODE18.HTM (2 of 10) [19/1/2003 1:35:18]

Lecture 18 - shortest path algorthms

We constructed a graph where the vertices were the possible words/positions in the sentence, with an edge between possible neighboring words. Listen To Part 20-2

file:///E|/LEC/LECTUR16/NODE18.HTM (3 of 10) [19/1/2003 1:35:18]

Lecture 18 - shortest path algorthms

The weight of each edge is a function of the probability that these two words will be next to each other in a sentence. `hive me' would be less than `give me', for example. The final system worked extremely well - identifying over 99% of characters correctly based on grammatical and statistical constraints. Dynamic programming (the Viterbi algorithm) can be used on the sentences to obtain the same results, by finding the shortest paths in the underlying DAG. Listen To Part 20-3

Finding Shortest Paths In an unweighted graph, the cost of a path is just the number of edges on the shortest path, which can be found in O(n+m) time via breadth-first search. In a weighted graph, the weight of a path between two vertices is the sum of the weights of the edges on a path. BFS will not work on weighted graphs because sometimes visiting more edges can lead to shorter distance, ie. 1+1+1+1+1+1+1 < 10. Note that there can be an exponential number of shortest paths between two nodes - so we cannot report all shortest paths efficiently. Note that negative cost cycles render the problem of finding the shortest path meaningless, since you can always loop around the negative cost cycle more to reduce the cost of the path.

file:///E|/LEC/LECTUR16/NODE18.HTM (4 of 10) [19/1/2003 1:35:18]

Lecture 18 - shortest path algorthms

Thus in our discussions, we will assume that all edge weights are positive. Other algorithms deal correctly with negative cost edges. Minimum spanning trees are uneffected by negative cost edges. Listen To Part 20-4

Dijkstra's Algorithm We can use Dijkstra's algorithm to find the shortest path between any two vertices and t in G. The principle behind Dijkstra's algorithm is that if had better be the shortest path from to x.

is the shortest path from to t, then

This suggests a dynamic programming-like strategy, where we store the distance from to all nearby nodes, and use them to find the shortest path to more distant nodes. The shortest path from to , d(,)=0. If all edge weights are positive, the smallest edge incident to , say (,x), defines d(,x). We can use an array to store the length of the shortest path to each node. Initialize each to

to start.

Soon as we establish the shortest path from to a new node x, we go through each of its incident edges to see if there is a better way from to other nodes thru x. Listen To Part 20-5

for i=1 to n,

for each edge (,v), dist[v]=d(,v)

last=

while (

)

file:///E|/LEC/LECTUR16/NODE18.HTM (5 of 10) [19/1/2003 1:35:18]

Lecture 18 - shortest path algorthms

select v such that

for each (v,x),

last=v

Complexity

if we use adjacency lists and a Boolean array to mark what is known.

This is essentially the same as Prim's algorithm. An implementation of Dijkstra's algorithm would be faster for sparse graphs, and comes from using a heap of the vertices (ordered by distance), and updating the distance to each vertex (if necessary) in time for each edge out from freshly known vertices. follows from using Fibonacci heaps, since they permit one to do a decrease-key Even better, operation in O(1) amortized time. Listen To Part 20-8

All-Pairs Shortest Path Notice that finding the shortest path between a pair of vertices (,t) in worst case requires first finding the shortest path from to all other vertices in the graph. Many applications, such as finding the center or diameter of a graph, require finding the shortest path between all pairs of vertices. We can run Dijkstra's algorithm n times (once from each possible start vertex) to solve all-pairs shortest path problem in . Can we do better? Improving the complexity is an open question but there is a super-slick dynamic programming algorithm

file:///E|/LEC/LECTUR16/NODE18.HTM (6 of 10) [19/1/2003 1:35:18]

Lecture 18 - shortest path algorthms

which also runs in

.

Listen To Part 20-9

Dynamic Programming and Shortest Paths The four-step approach to dynamic programming is: 1. 2. 3. 4.

Characterize the structure of an optimal solution. Recursively define the value of an optimal solution. Compute this recurrence in a bottom-up fashion. Extract the optimal solution from computed information.

From the adjacency matrix, we can construct the following matrix:

, if

D[i,j] = w(i,j),

and

is not in E

if

D[i,j] = 0, if i=j This tells us the shortest path going through no intermediate nodes. There are several ways to characterize the shortest path between two nodes in a graph. Note that the shortest path from i to j, , using at most M edges consists of the shortest path from i to k using at most M-1 edges + W(k, j) for some k. Listen To Part 20-10 This suggests that we can compute all-pair shortest path with an induction based on the number of edges in the optimal path. Let What is

be the length of the shortest path from i to j using at most m edges. ?

file:///E|/LEC/LECTUR16/NODE18.HTM (7 of 10) [19/1/2003 1:35:18]

Lecture 18 - shortest path algorthms

What if we know

for all i,j?

since w[k, k]=0 This gives us a recurrence, which we can evaluate in a bottom up fashion:

for i=1 to n

for j=1 to n

for k=1 to n

=Min(

This is an

,

)

algorithm just like matrix multiplication, but it only goes from m to m+1 edges.

Listen To Part 20-11 Since the shortest path between any two nodes must use at most n edges (unless we have negative cost cycles), we must repeat that procedure n times (m=1 to n) for an algorithm. We can improve this to with the observation that any path using at most 2m edges is the . So a function of paths using at most m edges each. This is just like computing logarithmic number of multiplications suffice for exponentiation.

file:///E|/LEC/LECTUR16/NODE18.HTM (8 of 10) [19/1/2003 1:35:18]

Lecture 18 - shortest path algorthms

Although this is slick, observe that even from each vertex!

is slower than running Dijkstra's algorithm starting

Listen To Part 20-12

The Floyd-Warshall Algorithm

An alternate recurrence yields a more efficient dynamic programming formulation. Number the vertices from 1 to n. Let be the shortest path from i to j using only vertices from 1, 2,..., k as possible intermediate vertices. What is

? With no intermediate vertices, any path consists of at most one edge, so

In general, adding a new vertex k+1 helps iff a path goes through it, so

Although this looks similar to the previous recurrence, it isn't. The following algorithm implements it:

for k=1 to n

for i=1 to n

for j=1 to n

file:///E|/LEC/LECTUR16/NODE18.HTM (9 of 10) [19/1/2003 1:35:18]

.

Lecture 18 - shortest path algorthms

This obviously runs in time, which asymptotically is no better than a calls to Dijkstra's algorithm. However, the loops are so tight and it is so short and simple that it runs better in practice by a constant factor.

Next: Lecture 19 - satisfiability Up: No Title Previous: Lecture 17 - minimum Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR16/NODE18.HTM (10 of 10) [19/1/2003 1:35:18]

Lecture 19 - satisfiability

Next: Lecture 20 - integer Up: No Title Previous: Lecture 18 - shortest

Lecture 19 - satisfiability Listen To Part 21-7

The Theory of NP-Completeness Several times this semester we have encountered problems for which we couldn't find efficient algorithms, such as the traveling salesman problem. We also couldn't prove an exponential time lower bound for the problem. By the early 1970s, literally hundreds of problems were stuck in this limbo. The theory of NP-Compleness, developed by Stephen Cook and Richard Karp, provided the tools to show that all of these problems were really the same problem. Listen To Part 21-8

Polynomial vs. Exponential Time

f(n) = n!

n

f(n) = n

10

0.01 s 0.1 s

1

20

0.02 s 0.4 s

1 ms

30

0.03

1 sec

40

0.04 s 1.6 s

50

0.05

100

0.1

s 0.9 s s 2.5 s s 10 s

s

3.63 ms 77.1 years years

18.3 min 13 days years

1,000 1.00 s 1 ms Listen To Part 21-9

The Main Idea Suppose I gave you the following algorithm to solve the bandersnatch problem: file:///E|/LEC/LECTUR16/NODE19.HTM (1 of 7) [19/1/2003 1:35:22]

Lecture 19 - satisfiability

Bandersnatch(G)

Convert G to an instance of the Bo-billy problem Y.

Call the subroutine Bo-billy on Y to solve this instance.

Return the answer of Bo-billy(Y) as the answer to G.

Such a translation from instances of one type of problem to instances of another type such that answers are preserved is called a reduction. Now suppose my reduction translates G to Y in O(P(n)): 1. If my Bo-billy subroutine ran in O(P'(n)) I can solve the Bandersnatch problem in O(P(n)+P'(n)) 2. If I know that is a lower-bound to compute Bandersnatch, then must be a lower-bound to compute Bo-billy. The second argument is the idea we use to prove problems hard! Listen To Part 21-10

Convex Hull and Sorting A nice example of a reduction goes from sorting numbers to the convex hull problem:

We must translate each number to a point. We can map x to

file:///E|/LEC/LECTUR16/NODE19.HTM (2 of 7) [19/1/2003 1:35:22]

.

Lecture 19 - satisfiability

Why? That means each integer is mapped to a point on the parabola

.

Listen To Part 21-11 Since this parabola is convex, every point is on the convex hull. Further since neighboring points on the convex hull have neighboring x values, the convex hull returns the points sorted by x-coordinate, ie. the original numbers.

Sort(S)

For each

, create point

.

Call subroutine convex-hull on this point set.

From the leftmost point in the hull,

read off the points from left to right.

Creating and reading off the points takes O(n) time. What does this mean? Recall the sorting lower bound of . If we could do convex hull in better than , we could sort faster than - which violates our lower bound. Thus convex hull must take Observe that any algorithm as well.

as well!!!

convex hull algorithm also gives us a complicated but correct

file:///E|/LEC/LECTUR16/NODE19.HTM (3 of 7) [19/1/2003 1:35:22]

sorting

Lecture 19 - satisfiability

Listen To Part 22-2

What is a problem? A problem is a general question, with parameters for the input and conditions on what is a satisfactory answer or solution. An instance is a problem with the input parameters specified. Example: The Traveling Salesman Problem: Given a weighted graph G, what tour Instance:

,

Solution:

cost= 27

,

minimizes ,

,

. ,

A problem with answers restricted to yes and no is called a decision problem. Most interesting optimization problems can be phrased as decision problems which capture the essence of the computation. Listen To Part 22-3 Example: The Traveling Salesman Decision Problem. Given a weighted graph G and integer k, does there exist a traveling salesman tour with cost

k?

Using binary search and the decision version of the problem we can find the optimal TSP solution. For convenience, from now on we will talk only about decision problems. Note that there are many possible ways to encode the input graph: adjacency matrices, edge lists, etc. All reasonable encodings will be within polynomial size of each other. The fact that we can ignore minor differences in encoding is important. We are concerned with the difference

file:///E|/LEC/LECTUR16/NODE19.HTM (4 of 7) [19/1/2003 1:35:22]

Lecture 19 - satisfiability

between algorithms which are polynomial and exponential in the size of the input. Listen To Part 22-4

Satisfiability Consider the following logic problem: Instance: A set V of variables and a set of clauses C over V. Question: Does there exist a satisfying truth assignment for C? Example 1:

and

A clause is satisfied when at least one literal in it is TRUE. C is satisfied when Example 2:

TRUE.

,

Although you try, and you try, and you try and you try, you can get no satisfaction. There is no satisfying assigment since must be FALSE (third clause), so but then the first clause is unsatisfiable!

must be FALSE (second clause),

For various reasons, it is known that satisfiability is a hard problem. Every top-notch algorithm expert in the world (and countless other, lesser lights) have tried to come up with a fast algorithm to test whether a given set of clauses is satisfiable, but all have failed. Listen To Part 22-5 Further, many strange and impossible-to-believe things have been shown to be true if someone in fact did find a fast satisfiability algorithm. Clearly, Satisfiability is in NP, since we can guess an assignment of TRUE, FALSE to the literals and check it in polynomial time. Listen To Part 22-10

P versus NP The precise distinction between whether a problem is in P or NP is somewhat technical, requiring formal file:///E|/LEC/LECTUR16/NODE19.HTM (5 of 7) [19/1/2003 1:35:22]

Lecture 19 - satisfiability

language theory and Turing machines to state correctly. However, intuitively a problem is in P, (ie. polynomial) if it can be solved in time polynomial in the size of the input. A problem is in NP if, given the answer, it is possible to verify that the answer is correct within time polynomial in the size of the input. Example P - Is there a path from to t in G of length less than k. Example NP - Is there a TSP tour in G of length less than k. Given the tour, it is easy to add up the costs and convince me it is correct. Example not NP - How many TSP tours are there in G of length less than k. Since there can be an exponential number of them, we cannot count them all in polynomial time. Don't let this issue confuse you - the important idea here is of reductions as a way of proving hardness. Listen To Part 22-7

3-Satisfiability Instance: A collection of clause C where each clause contains exactly 3 literals, boolean variable v. Question: Is there a truth assignment to v so that each clause is satisfied? Note that this is a more restricted problem than SAT. If 3-SAT is NP-complete, it implies SAT is NP-complete but not visa-versa, perhaps long clauses are what makes SAT difficult?! After all, 1-Sat is trivial! Theorem: 3-SAT is NP-Complete Proof: 3-SAT is NP - given an assignment, just check that each clause is covered. To prove it is complete, a reduction from must be provided. We will transform each clause independantly based on its length. Suppose the clause ●

contains k literals.

If k=1, meaning ,

, create two new variables ,

,

.

file:///E|/LEC/LECTUR16/NODE19.HTM (6 of 7) [19/1/2003 1:35:22]

and four new 3-literal clauses:

Lecture 19 - satisfiability

Note that the only way all four of these can be satisfied is if z is TRUE.

● ● ●

Listen To Part 22-8 If k=2, meaning If k=3, meaning If k>3, meaning

, create one new variable and two new clauses: , , copy into the 3-SAT instance as it is. , create n-3 new variables and n-2 new clauses in a chain:

, ...

If none of the original variables in a clause are TRUE, there is no way to satisfy all of them using the additional variable:

But if any literal is TRUE, we have n-3 free variables and n-3 remaining 3-clauses, so we can satisfy each of them. Since any SAT solution will also satisfy the 3-SAT instance and any 3-SAT solution sets variables giving a SAT solution - the problems are equivallent. If there were n clauses and m total literals in the SAT instance, this transform takes O(m) time, so SAT and 3-SAT. Note that a slight modification to this construction would prove 4-SAT, or 5-SAT,... also NP-complete. However, it breaks down when we try to use it for 2-SAT, since there is no way to stuff anything into the chain of clauses. It turns out that resolution gives a polynomial time algorithm for 2-SAT. Listen To Part 22-9 Having at least 3-literals per clause is what makes the problem difficult. Now that we have shown 3-SAT is NP-complete, we may use it for further reductions. Since the set of 3-SAT instances is smaller and more regular than the SAT instances, it will be easier to use 3-SAT for future reductions. Remember the direction to reduction!

Next: Lecture 20 - integer Up: No Title Previous: Lecture 18 - shortest Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR16/NODE19.HTM (7 of 7) [19/1/2003 1:35:22]

Lecture 20 - integer programming

Next: Lecture 21 - vertex Up: No Title Previous: Lecture 19 - satisfiability

Lecture 20 - integer programming Listen To Part 22-6 36.4-5 Give a polynomial-time algorithm to satisfy Boolean formulas in disjunctive normal form.

Satisfying one clause in DFS satisfied the whole formula. One clause can always be satisfied iff it does not contain both a variable and its complement. Why not use this reduction to give a polynomial-time algorithm for 3-SAT? The DNF formula can become exponentially large and hence the reduction cannot be done in polynomial time. Listen To Part 24-2

A Perpetual Point of Confusion Note carefully the direction of the reduction. We must transform every instance of a known NP-complete problem to an instance of the problem we are interested in. If we do the reduction the other way, all we get is a slow way to solve x, by using a subroutine which probably will take exponential time. This always is confusing at first - it seems bass-ackwards. Make sure you understand the direction of reduction now - and think back to this when you get confused. Listen To Part 24-3

Integer Programming Instance: A set v of integer variables, a set of inequalities over these variables, a function f(v) to maximize, and integer B. Question: Does there exist an assignment of integers to v such that all inequalities are true and

file:///E|/LEC/LECTUR16/NODE20.HTM (1 of 4) [19/1/2003 1:35:24]

?

Lecture 20 - integer programming

Example:

A solution to this is

,

.

Example:

Since the maximum value of f(v) given the constraints is

, there is no solution.

Theorem: Integer Programming is NP-Hard Proof: By reduction from Satisfiability Any set instance has boolean variables and clauses. Our Integer programming problem will have twice as many variables as the SAT instance, one for each variable and its compliment, as well as the following inequalities: Listen To Part 24-4 For each variable ●

in the set problem, we will add the following constraints: and

Both IP variables are restricted to values of 0 or 1, which makes them equivalent to boolean variables restricted to true/false. ●

Exactly one of the IP variables associated with a given sat variable is 1. This means that exactly one of and are true! file:///E|/LEC/LECTUR16/NODE20.HTM (2 of 4) [19/1/2003 1:35:24]

Lecture 20 - integer programming ●

for each clause

in the sat instance, construct a constraint:

Thus at least one IP variable must be one in each clause! Thus satisfying the constraint is equivalent to satisfying the clause! Our maximization function and bound are relatively unimportant:

B=0.

Clearly this reduction can be done in polynomial time. Listen To Part 24-5 We must show: 1. Any SAT solution gives a solution to the IP problem. In any SAT solution, a TRUE literal corresponds to a 1 in the IP, since if the expression is SATISFIED, at least one literal per clause in TRUE, so the sum in the inequality is 1. 2. Any IP solution gives a SAT solution. Given a solution to this IP instance, all variables will be 0 or 1. Set the literals correspondly to 1 variable TRUE and the 0 to FALSE. No boolean variable and its complement will both be true, so it is a legal assignment with also must satisfy the clauses. Neat, sweet, and NP-complete! Listen To Part 24-6

Things to Notice 1. The reduction preserved the structure of the problem. Note that the reduction did not solve the problem - it just put it in a different format. 2. The possible IP instances which result are a small subset of the possible IP instances, but since some of them are hard, the problem in general must be hard. 3. The transformation captures the essence of why IP is hard - it has nothing to do with big coefficients or big ranges on variables; for restricting to 0/1 is enough. A careful study of what properties we do need for our reduction tells us a lot about the problem. 4. It is not obvious that IP NP, since the numbers assigned to the variables may be too large to write in polynomial time - don't be too hasty!

file:///E|/LEC/LECTUR16/NODE20.HTM (3 of 4) [19/1/2003 1:35:24]

Lecture 20 - integer programming

Next: Lecture 21 - vertex Up: No Title Previous: Lecture 19 - satisfiability Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR16/NODE20.HTM (4 of 4) [19/1/2003 1:35:24]

Lecture 21 - vertex cover

Next: Lecture 22 - techniques Up: No Title Previous: Lecture 20 - integer

Lecture 21 - vertex cover Listen To Part 24-7 36.5-2 - Given an integer matrix A, and in integer m-vector b, the 0-1 integer programming problem asks whether there is an integer n-vector x with elements in the set (0,1) such that . Prove that 0-1 integer programming is NP-hard (hint: reduce from 3-SAT).

This is really the exact same problem as the previous integer programming problem, slightly concealed by: ● ●

The linear algebra notation - each row is one constraint. All inequalities are - multiply both sides by -1 to reverse the constraint from

to

if necessary.

Listen To Part 24-8

Vertex Cover Instance: A graph G=(V, E), and integer Question: Is there a subset of at most k vertices such that every

has at least one vertex in the subset?

Here, four of the eight vertices are enough to cover. It is trivial to find a vertex cover of a graph - just take all the vertices. The tricky part is to cover with as small a set as possible. Theorem: Vertex cover is NP-complete. file:///E|/LEC/LECTUR16/NODE21.HTM (1 of 7) [19/1/2003 1:35:28]

Lecture 21 - vertex cover

Proof: VC in in NP - guess a subset of vertices, count them, and show that each edge is covered. To prove completeness, we show 3-SAT and VC. From a 3-SAT instance with n variables and C clauses, we construct a graph with 2N+3C vertices. Listen To Part 24-9 For each variable, we create two vertices connected by an edge:

To cover each of these edges, at least n vertices must be in the cover, one for each pair. For each clause, we create three new vertices, one for each literal in each clause. Connect these in a triangle. At least two vertices per triangle must be in the cover to take care of edges in the triangle, for a total of at least 2C vertices. Finally, we will connect each literal in the flat structure to the corresponding vertices in the triangles which share the same literal.

Listen To Part 24-10 Claim: This graph will have a vertex cover of size N+2C if and only if the expression is satisfiable. By the earlier analysis, any cover must have at least N+2C vertices. To show that our reduction is correct, we must show that: 1. Every satisfying truth assignment gives a cover. Select the N vertices cooresponding to the TRUE literals to be in the cover. Since it is a satisfying truth assignment, at least one of the three cross edges associated with each clause must already be covered - pick the other two vertices to complete the cover. 2. Every vertex cover gives a satisfying truth assignment.

file:///E|/LEC/LECTUR16/NODE21.HTM (2 of 7) [19/1/2003 1:35:28]

Lecture 21 - vertex cover

Every vertex cover must contain n first stage vertices and 2C second stage vertices. Let the first stage vertices define the truth assignment. To give the cover, at least one cross-edge must be covered, so the truth assignment satisfies. For a cover to have N+2C vertices, all the cross edges must be incident on a selected vertex. Let the N selected vertices from the first stage coorespond to TRUE literals. If there is a satisfying truth assignment, that means at least one of the three cross edges from each triangle is incident on a TRUE vertex. By adding the other two vertices to the cover, we cover all edges associated with the clause. Every SAT defines a cover and Every Cover Truth values for the SAT! Example:

,

.

Listen To Part 25-1

Starting from the Right Problem As you can see, the reductions can be very clever and very complicated. While theoretically any NPcomplete problem can be reduced to any other one, choosing the correct one makes finding a reduction much easier.

As you can see, the reductions can be very clever and complicated. While theoretically any NP-complete problem will do, choosing the correct one can make it much easier.

Maximum Clique Instance: A graph G=(V,E) and integer

.

file:///E|/LEC/LECTUR16/NODE21.HTM (3 of 7) [19/1/2003 1:35:28]

Lecture 21 - vertex cover

Question: Does the graph contain a clique of j vertices, ie. is there a subset of v of size j such that every pair of vertices in the subset defines an edge of G? Example: this graph contains a clique of size 5.

Listen To Part 25-2 When talking about graph problems, it is most natural to work from a graph problem - the only NPcomplete one we have is vertex cover! Theorem: Clique is NP-complete Proof: If you take a graph and find its vertex cover, the remaining vertices form an independent set, meaning there are no edges between any two vertices in the independent set, for if there were such an edge the rest of the vertices could not be a vertex cover.

Clearly the smallest vertex cover gives the biggest independent set, and so the problems are equivallent Delete the subset of vertices in one from the total set of vertices to get the order! Thus finding the maximum independent set must be NP-complete! Listen To Part 25-3 In an independent set, there are no edges between two vertices. In a clique, there are always between two vertices. Thus if we complement a graph (have an edge iff there was no edge in the original graph), a clique becomes an independent set and an independent set becomes a Clique! file:///E|/LEC/LECTUR16/NODE21.HTM (4 of 7) [19/1/2003 1:35:28]

Lecture 21 - vertex cover

Thus finding the largest clique is NP-complete: If VC is a vertex cover in G, then V-VC is a clique in G'. If C is a clique in G, V-C is a vertex cover in G'. Listen To Part 25-4 36.5-1 Prove that subgraph isomorphism is NP-complete.

1. Guessing a subgraph of G and proving it is isomorphism to h takes time, so it is in NP. 2. Clique and subgraph isomorphism. We must transform all instances of clique into some instances of subgraph isomorphism. Clique is a special case of subgraph isomorphism! Thus the following reduction suffices. Let G=G' and

, the complete subgraph on k nodes.

Listen To Part 25-5

Integer Partition (Subset Sum) Instance: A set of integers S and a target integer t. Problem: Is there a subset of S which adds up exactly to t? Example:

and T=3754

Answer: 1+16+64+256+1040+1093+1284 = T Observe that integer partition is a number problem, as opposed to the graph and logic problems we have file:///E|/LEC/LECTUR16/NODE21.HTM (5 of 7) [19/1/2003 1:35:28]

Lecture 21 - vertex cover

seen to date. Theorem: Integer Partition is NP-complete. Proof: First, we note that integer partition is in NP. Guess a subset of the input number and simply add them up. To prove completeness, we show that vertex cover incidence matrix to represent the graph G.

integer partition. We use a data structure called an

Listen To Part 25-6 How many 1's are there in each column? Exactly two. How many 1's in a particular row? Depends on the vertex degree. The reduction from vertex cover will create n+m numbers from G. The numbers from the vertices will be a base-4 realization of rows from the incidence matrix, plus a high order digit:

ie.

becomes

The numbers from the edges will be

. .

The target integer will be

file:///E|/LEC/LECTUR16/NODE21.HTM (6 of 7) [19/1/2003 1:35:28]

Lecture 21 - vertex cover

Why? Each column (digit) represents an edge. We want a subset of vertices which covers each edge. We can only use k x vertex/numbers, because of the high order digit of the target.

Listen To Part 25-7 We might get only one instance of each edge in a cover - but we are free to take extra edge/numbers to grab an extra 1 per column.

VC in G

Integer Partition in S

Given k vertices covering G, pick the k cooresponding vertex/numbers. Each edge in G is incident on one or two cover vertices. If it is one, includes the cooresponding edge/number to give two per column.

Integer Partition in S

VC in G

Any solution to S must contain exactly k vertex/numbers. Why? It cannot be more because the target in that digit is k and it cannot be less because, with at most 3 1's per edge/digit-column, no sum of these can carry over into the next column. (This is why base-4 number were chosen). This subset of k vertex/numbers must contain at least one edge-list per column, since if not there is no way to account for the two in each column of the target integer, given that we can pick up at most one edge-list using the edge number. (Again, the prevention of carrys across digits prevents any other possibilites). Neat, sweet, and NP-complete! Notice that this reduction could not be performed in polynomial time if the number were written in unary 5=11111. Big numbers is what makes integer partition hard!

Next: Lecture 22 - techniques Up: No Title Previous: Lecture 20 - integer Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR16/NODE21.HTM (7 of 7) [19/1/2003 1:35:28]

Lecture 22 - techniques for proving hardness

Next: Lecture 23 - approximation Up: No Title Previous: Lecture 21 - vertex

Lecture 22 - techniques for proving hardness Hamiltonian Cycle Instance: A graph G Question: Does the graph contains a HC, i.e. an ordered of the vertices

?

This problem is intimately relates to the Traveling Salesman. Question: Is there an ordering of the vertices of a weighted graph such that

?

. Assign each edge in G weight 1, any edge not in G weight 2. This new graph has a Clearly, Traveling Salesman tour of cost n iff the graph is Hamiltonian. Thus TSP is NP-complete if we can show HC is NP-complete. Theorem: Hamiltonian Circuit is NP-complete Proof: Clearly HC is in NP-guess a permutation and check it out. To show it is complete, we use vertex cover. A vertex cover instance consists of a graph and a constant k, the minimum size of an acceptable cover. We must construct another graph. Each edge in the initial graph will be represented by the following component:

All further connections to this gadget will be through vertices , , and . The key observation about this gadget is that there are only three ways to traverse all the vertices:

file:///E|/LEC/LECTUR16/NODE22.HTM (1 of 7) [19/1/2003 1:35:31]

Lecture 22 - techniques for proving hardness

Note that in each case, we exit out the same side we entered. Each side of each edge gadget is associated with a vertex. Assuming some arbitrary order to the edges incident on a particular vertex, we can link successive gadgets by edges forming a chain of gadgets. Doing this for all vertices in the original graph creates n intertwined chains with n entry points and n exits.

Thus we have encoded the information about the initial graph. What about k? We set up k additional vertices and connect each of these to the n start points and n end points of each chain.

Total size of new graph: GE+K vertices and 12E+2kN+2E edges and time.

construction is polynomial in size

We claim this graph has a HC iff G has a VC of size k. 1. Suppose

is a HC.

Assume it starts at one of the k selector vertices. It must then go through one of the chains of gadgets until it reaches a different selector vertex. Since the tour is a HC, all gadgets are traversed. The k chains correspond to the vertices in the cover. Note that if both vertices associated with an edge are in the cover, the gadget will be traversal in

file:///E|/LEC/LECTUR16/NODE22.HTM (2 of 7) [19/1/2003 1:35:31]

Lecture 22 - techniques for proving hardness

two pieces - otherwise one chain suffices. To avoid visiting a vertex more than once, each chain is associated with a selector vertex. 2. Now suppose we have a vertex cover of size . We can always add more vertices to the cover to bring it up to size k. For each vertex in the cover, start traversing the chain. At each entry point to a gadget, check if the other vertex is in the cover and traverse the gadget accordingly. Select the selector edges to complete the circuit. Neat, sweet, and NP-complete. To show that Longest Path or Hamiltonian Path is NP-complete, add start and stop vertices and distinguish the first and last selector vertices.

This has a Hamiltonian path from start to stop iff the original graph has a vertex cover of size k. Listen To Part 26-2

Other NP-complete Problems ● ● ● ● ●

Partition - can you partition n integers into two subsets so that the sums of the subset are equal? Bin Packing - how many bins of a given size do you need to hold n items of variable size? Chromatic Number - how many colors do you need to color a graph? checkers - does black have a forced win from a given position? Scheduling, Code Optimization, Permanent Evaluation, Quadratic Programming, etc.

Open: Graph Isomorphism, Composite Number, Minimum Length Triangulation. Listen To Part 26-3

Polynomial or Exponential?

file:///E|/LEC/LECTUR16/NODE22.HTM (3 of 7) [19/1/2003 1:35:31]

Lecture 22 - techniques for proving hardness

Just changing a problem a little can make the difference between it being in P or NP-complete: P

NP-complete

Shortest Path

Longest Path

Eulerian Circuit Hamiltonian Circuit Edge Cover

Vertex Cover

The first thing you should do when you suspect a problem might be NP-complete is look in Garey and Johnson, Computers and Intractability. It contains a list of several hundred problems known to be NPcomplete. Either what you are looking for will be there or you might find a closely related problem to use in a reduction. Listen To Part 26-4

Techniques for Proving NP-completeness 1. Restriction - Show that a special case of the problem you are interested in is NP-complete. For example, the problem of finding a path of length k is really Hamiltonian Path. 2. Local Replacement - Make local changes to the structure. An example is the reduction . Another example is showing isomorphism is no easier for bipartite graphs:

For any graph, replacing an edge with makes it bipartite. 3. Component Design - These are the ugly, elaborate constructions Listen To Part 26-5

The Art of Proving Hardness Proving that problems are hard is an skill. Once you get the hang of it, it is surprisingly straightforward and pleasurable to do. Indeed, the dirty little secret of NP-completeness proofs is that they are usually easier to recreate than explain, in the same way that it is usually easier to rewrite old code than the try to understand it.

file:///E|/LEC/LECTUR16/NODE22.HTM (4 of 7) [19/1/2003 1:35:31]

Lecture 22 - techniques for proving hardness

I offer the following advice to those needing to prove the hardness of a given problem: ●

Make your source problem as simple (i.e. restricted) as possible.

●

Never use the general traveling salesman problem (TSP) as a target problem. Instead, use TSP on instances restricted to the triangle inequality. Better, use Hamiltonian cycle, i.e. where all the weights are 1 or . Even better, use Hamiltonian path instead of cycle. Best of all, use Hamiltonian path on directed, planar graphs where each vertex has total degree 3. All of these problems are equally hard, and the more you can restrict the problem you are reducing, the less work your reduction has to do. Make your target problem as hard as possible. Don't be afraid to add extra constraints or freedoms in order to make your problem more general (at least temporarily).

●

Listen To Part 26-6 Select the right source problem for the right reason. Selecting the right source problem makes a big difference is how difficult it is to prove a problem hard. This is the first and easiest place to go wrong. I usually consider four and only four problems as candidates for my hard source problem. Limiting them to four means that I know a lot about these problems - which variants of these problems are hard and which are soft. My favorites are: ❍

❍

❍

❍

3-Sat - that old reliable... When none of the three problems below seem appropriate, I go back to the source. Integer partition - the one and only choice for problems whose hardness seems to require using large numbers. Vertex cover - for any graph problems whose hardness depends upon selection. Chromatic number, clique, and independent set all involve trying to select the correct subset of vertices or edges. Hamiltonian path - for any graph problems whose hardness depends upon ordering. If you are trying to route or schedule something, this is likely your lever.

Listen To Part 26-7 ●

Amplify the penalties for making the undesired transition. You are trying to translate one problem into another, while making them stay the same as much as possible. The easiest way to do this is to be bold with your penalties, to punish anyone trying to deviate from your proposed solution. ``If you pick this, then you have to pick up this huge set

file:///E|/LEC/LECTUR16/NODE22.HTM (5 of 7) [19/1/2003 1:35:31]

Lecture 22 - techniques for proving hardness

●

which dooms you to lose.'' The sharper the consequences for doing what is undesired, the easier it is to prove if and only if. Think strategically at a high level, then build gadgets to enforce tactics.

●

You should be asking these kinds of questions. ``How can I force that either A or B but not both are chosen?'' ``How can I force that A is taken before B?'' ``How can I clean up the things I did not select?'' Alternate between looking for an algorithm or a reduction if you get stuck. Sometimes the reason you cannot prove hardness is that there is an efficient algorithm to solve your problem! When you can't prove hardness, it likely pays to change your thinking at least for a little while to keep you honest.

Listen To Part 26-8

Now watch me try it! To demonstrate how one goes about proving a problem hard, I accept the challenge of showing how a proof can be built on the fly. I need a volunteer to pick a random problem from the 400+ hard problems in the back of Garey and Johnson. Listen To Part 27-2

Dealing with NP-complete Problems

Option 1: Algorithm fast in the Average case Examples are Branch-and-bound for the Traveling Salesman Problem, backtracking algorithms, etc.

Option 2: Heuristics Heuristics are rules of thumb; fast methods to find a solution with no requirement that it be the best one. Note that the theory of NP-completeness does not stipulate that it is hard to get close to the answer, only that it is hard to get the optimal answer.

file:///E|/LEC/LECTUR16/NODE22.HTM (6 of 7) [19/1/2003 1:35:31]

Lecture 22 - techniques for proving hardness

Often, we can prove performance bounds on heuristics, that the resulting answer is within C times that of the optimal one.

Next: Lecture 23 - approximation Up: No Title Previous: Lecture 21 - vertex Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR16/NODE22.HTM (7 of 7) [19/1/2003 1:35:31]

Lecture 23 - approximation algorithms and Cook's theorem

Next: Index Up: No Title Previous: Lecture 22 - techniques

Lecture 23 - approximation algorithms and Cook's theorem Listen To Part 26-1 36.5-5 Prove that Hamiltonian Path is NP-complete.

This is not a special case of Hamiltonian cycle! (G may have a HP but not cycle) The easiest argument says that G contains a HP but no HC iff (x,y) in G such that adding edge (x, y) to G causes to have a HC, so calls to a HC function solves HP. The cleanest proof modifies the VC and HC reduction from the book:

Listen To Part 27-3

Approximating Vertex Cover As we have seen, finding the minimum vertex cover is NP-complete. However, a very simple strategy (heuristic) can get us a cover at most twice that of the optimal.

While the graph has edges

pick an arbitrary edge v, u file:///E|/LEC/LECTUR16/NODE23.HTM (1 of 11) [19/1/2003 1:35:35]

Lecture 23 - approximation algorithms and Cook's theorem

add both u and v to the cover

delete all edges incident on either u and v

If the graph is represented by an adjacency list this can be implemented in O(m+n) time. This heuristic must always produce cover, since an edge is only deleted when it is adjacent to a cover vertex. Further, any cover uses at least half as many vertices as the greedy cover.

Why? Delete all edges from the graph except the edges we selected. No two of these edges share a vertex. Therefore, any cover of just these edges must include one vertex per edge, or half the greedy cover! Listen To Part 27-4

Things to Notice ●

Although the heuristic is simple, it is not stupid. Many other seemingly smarter ones can give a far worse performance in the worst case. Example: Pick one of the two vertices instead of both (after all, the middle edge is already covered) The optimal cover is one vertex, the greedy heuristic is two vertices, while the new/bad heuristic can be as bad as n-1.

file:///E|/LEC/LECTUR16/NODE23.HTM (2 of 11) [19/1/2003 1:35:35]

Lecture 23 - approximation algorithms and Cook's theorem

● ●

●

Proving a lower bound on the optimal solution is the key to getting an approximation result. Making a heuristic more complicated does not necessarily make it better. It just makes it more difficult to analyze. A post-processing clean-up step (delete any unecessessary vertex) can only improve things in practice, but might not help the bound.

Listen To Part 27-5

The Euclidean Traveling Salesman In the traditional version of TSP - a salesman wants to plan a drive to visit all his customers exactly once and get back home. Euclidean geometry satisfies the triangle inequality,

.

TSP remains hard even when the distances are Euclidean distances in the plane.

Note that the cost of airfares is an example of a distance function which violates the triangle inequality. However, we can approximate the optimal Euclidean TSP tour using minimum spanning trees. Claim: the cost of a MST is a lower bound on the cost of a TSP tour. Why? Deleting any edge from a TSP tour leaves a path, which is a tree of weight at least that of the MST! Listen To Part 27-6 If we were allowed to visit cities more than once, doing a depth-first traversal of a MST, and then walking out the tour specified is at most twice the cost of MST. Why? We will be using each edge exactly twice. file:///E|/LEC/LECTUR16/NODE23.HTM (3 of 11) [19/1/2003 1:35:35]

Lecture 23 - approximation algorithms and Cook's theorem

Every edge is used exactly twice in the DFS tour: 1. However, how can we avoid revisiting cities? We can take a shortest path to the next unvisited vertex. The improved tour is 1-2-3-5-8-9-6-4-7-10-11-1. Because we replaced a chain of edges by the edge, the triangle inequality ensures the tour only gets shorter. Thus this is still within twice optimal! Listen To Part 27-1 37.1-3 Give an efficient greedy algorithm that finds an optimal vertex cover of a tree in linear time.

In a vertex cover we need to have at least one vertex for each edge. Every tree has at least two leaves, meaning that there is always an edge which is adjacent to a leaf. Which vertex can we never go wrong picking? The non-leaf, since it is the only one which can also cover other edges! After trimming off the covered edges, we have a smaller tree. We can repeat the process until the tree as 0 or 1 edges. When the tree consists only of an isolated edge, pick either vertex. All leaves can be identified and trimmed in O(n) time during a DFS.

Formal Languages and the Theory of NP-completeness The theory of NP-completeness is based on formal languages and Turing machines, and so we will must work on a more abstract level than usual. For a given alphabet of symbols 0, 1, &, we can form an infinite set of strings or words by arranging them in any order: `&10', `111111',`&&&', and `&'.

file:///E|/LEC/LECTUR16/NODE23.HTM (4 of 11) [19/1/2003 1:35:35]

Lecture 23 - approximation algorithms and Cook's theorem

A subset of the set of strings over some alphabet is a formal language. Formal language theory concerns the study of how powerful a machine you need to recognize whether a string is from a particular language. Example: Is the string a binary representation of a even number? A simple finite machine can check if the last symbol is zero:

No memory is required, except for the current state. Observe that solving decision problems can be thought of as formal language recognition. The problem instances are encoded as strings and strings in the language if and only if the answer to the decision problem is YES! What kind of machine is necessary to recognize this language? A Turing Machine! A Turing machine has a finite-state-control (its program), a two way infinite tape (its memory) and a readwrite head (its program counter)

So, where are we? Each instance of an optimization or decision problem can be encoded as string on some alphabet. The set of all instances which return True for some problem define a language. Hence, any problem which solves this problem is equivalent to a machine which recognizes whether an instance is in the language! The goal of all this is going to be a formal way to talk about the set of problems which can be solved in file:///E|/LEC/LECTUR16/NODE23.HTM (5 of 11) [19/1/2003 1:35:35]

Lecture 23 - approximation algorithms and Cook's theorem

polynomial time, and the set that cannot be.

Non-deterministic Turing Machines Suppose we buy a guessing module peripherial for our Turing machine, which looks at a Turing machine program and problem instance and in polynomial time writes something it says is an answer. To convince ourselves it really is an answer, we can run another program to check it. Ex: The Traveling Salesman Problem The guessing module can easily write a permutation of the vertices in polynomial time. We can check if it is correct by summing up the weights of the special edges in the permutation and see that it is less than k.

The class of languages which we can recognize in time polynomial in the size of the string or a deterministic Turing Machine (without guessing module) is called P. The class of languages we can recognize in time polynomial in the length of the string or a nondeterministic Turing Machine is called NP. Clearly, , since for any DTM program we can run it on a non-deterministic machine, ignore what the guessing module is doing, and it will just as fast.

P ?= NP Observe that any NDTM program which takes time P(n) can simulated in time on a deterministic machine, by running the checking program times, once on each possible guessed string.

The $10,000 question is whether a polynomial time simulation exists, or in other words whether P=NP?. Do there exist languages which can be verified in polynomial time and still take exponential time on deterministic machines? This is the most important question in computer science. Since proving an exponential time lower bound for a problem in NP would make us famous, we assume that we cannot do it. file:///E|/LEC/LECTUR16/NODE23.HTM (6 of 11) [19/1/2003 1:35:35]

Lecture 23 - approximation algorithms and Cook's theorem

What we can do is prove that it is at least as hard as any problem in NP. A problem in NP for which a polynomial time algorithm would imply all languages in NP are in P is called NP-complete.

Turing Machines and Cook's Theorem Cook's Theorem proves that satisfiability is NP-complete by reducing all non-deterministic Turing machines to SAT. Each Turing machine has access to a two-way infinite tape (read/write) and a finite state control, which serves as the program.

A program for a non-deterministic TM is: 1. 2. 3. 4.

Space on the tape for guessing a solution and certificate to permit verification. A finite set of tape symbols A finite set of states for the machine, including the start state and final states A transition function, which takes the current machine state, and current tape symbol and returns the new state, symbol, and head position.

We know a problem is in NP if we have a NDTM program to solve it in worst-case time p[n], where p is a polynomial and n is the size of the input.

Cook's Theorem - Satisfiability is NP-complete! Proof: We must show that any problem in NP is at least as hard as SAT. Any problem in NP has a nondeterministic TM program which solves it in polynomial time, specifically P(n). We will take this program and create from it an instance of satisfiability such that it is satisfiable if and only if the input string was in the language.

file:///E|/LEC/LECTUR16/NODE23.HTM (7 of 11) [19/1/2003 1:35:35]

Lecture 23 - approximation algorithms and Cook's theorem

If a polynomial time transform exists, then SAT must be NP-complete, since a polynomial solution to SAT gives a polynomial time algorithm to anything in NP. Our transformation will use boolean variables to maintain the state of the TM: Variable Range

Intended meaning

Q[i, j]

At time i, M is in state At time i, the read-write head

H[i,j]

is scanning tape square j S[i,j,k]

At time i, the contents of tape square j is symbol

literals, a polynomial number if p(n) is polynomial.

Note that there are

We will now have to add clauses to ensure that these variables takes or the values as in the TM computation. The group 6 clauses enforce the transition function of the machine. If the read-write head is not on tape square j at time i, it doesn't change .... There are

literals and

clauses in all, so the transformation is done in polynomial time!

Polynomial Time Reductions A decision problem is NP-hard if the time complexity on a deterministic machine is within a polynomial factor of the complexity of any problem in NP. file:///E|/LEC/LECTUR16/NODE23.HTM (8 of 11) [19/1/2003 1:35:35]

Lecture 23 - approximation algorithms and Cook's theorem

A problem is NP-complete if it is NP-hard and in NP. Cook's theorem proved SATISFIABILITY was NPhard by using a polynomial time reduction translating each problem in NP into an instance of SAT:

Since a polynomial time algorithm for SAT would imply a polynomial time algorithm for everything in NP, SAT is NP-hard. Since we can guess a solution to SAT, it is in NP and thus NP-complete. The proof of Cook's Theorem, while quite clever, was certainly difficult and complicated. We had to show that all problems in NP could be reduced to SAT to make sure we didn't miss a hard one. But now that we have a known NP-complete problem in SAT. For any other problem, we can prove it NPhard by polynomially transforming SAT to it!

Since the composition of two polynomial time reductions can be done in polynomial time, all we need show is that SAT, ie. any instance of SAT can be translated to an instance of x in polynomial time. Listen To Part 27-7

Finding the Optimal Spouse

1. There are up to n possible candidates we will see over our lifetime, one at a time. 2. We seek to maximize our probability of getting the single best possible spouse. file:///E|/LEC/LECTUR16/NODE23.HTM (9 of 11) [19/1/2003 1:35:35]

Lecture 23 - approximation algorithms and Cook's theorem

3. Our assessment of each candidate is relative to what we have seen before. 4. We must decided either to marry or reject each candidate as we see them. There is no going back once we reject someone. 5. Each candidate is ranked from 1 to n, and all permutations are equally likely. Listen To Part 27-8 For example, if the input permutation is

we see (3,1,2) after three candidates. Picking the first or last candidate gives us a probability of 1/n of getting the best. Since we seek maximize our chances of getting the best, it never pays to pick someone who is not the best we have seen. The optimal strategy is clearly to sample some fraction of the candidates, then pick the first one who is better than the best we have seen. But what is the fraction? Listen To Part 27-9 For a given fraction 1/f, what is the probability of finding the best? Suppose i+1 is the highest ranked person in the first n/f candidates. We win whenever the best candidate occurs before any number from 2 to i in the last n (1- 1/f) / f candidates. There is a 1/i probability of that, so,

In fact, the optimal is obtained by sampling the first n/e candidates.

Next: Index Up: No Title Previous: Lecture 22 - techniques file:///E|/LEC/LECTUR16/NODE23.HTM (10 of 11) [19/1/2003 1:35:35]

Lecture 23 - approximation algorithms and Cook's theorem

Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR16/NODE23.HTM (11 of 11) [19/1/2003 1:35:35]

About this document ...

Up: No Title Previous: Index

About this document ... This document was generated using the LaTeX2HTML translator Version 96.1 (Feb 5, 1996) Copyright © 1993, 1994, 1995, 1996, Nikos Drakos, Computer Based Learning Unit, University of Leeds. The command line arguments were: latex2html all.tex. The translation was initiated by Algorithms on Mon Jun 2 09:21:39 EDT 1997

Algorithms Mon Jun 2 09:21:39 EDT 1997

file:///E|/LEC/LECTUR16/NODE25.HTM [19/1/2003 1:35:36]

1.1.6 Kd-Trees

1.1.6 Kd-Trees

INPUT

OUTPUT

Input Description: A set S of n points in k -dimensions. Problem: Construct a tree which partitions the space by half-planes such that each point is contained in it is own region.

Implementations ● ● ●

Ranger - Nearest Neighbor Search in Higher Dimensions (C) (rating 8) Handbook of Algorithms and Data Structures (Pascal) (rating 3) DIMACS Implementation Challenges (FORTRAN) (rating 1)

Related Problems ●

Nearest Neighbor Search file:///E|/WEBSITE/FILES/00000007.HTM (1 of 2) [19/1/2003 1:35:38]

1.1.6 Kd-Trees ● ●

Point Location Range Search

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/00000007.HTM (2 of 2) [19/1/2003 1:35:38]

1.1.3 Suffix Trees and Arrays

1.1.3 Suffix Trees and Arrays

INPUT

Implementations ● ●

Stony Brook Project Implementations (C++) (rating 6) Handbook of Algorithms and Data Structures (Pascal) (rating 2)

Related Problems ● ● ●

Longest Common Substring Shortest Common Superstring String Matching

Go to the corresponding chapter in the book About the Book file:///E|/WEBSITE/FILES/SUFFREES.HTM (1 of 2) [19/1/2003 1:35:39]

OUTPUT

1.1.3 Suffix Trees and Arrays

Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/SUFFREES.HTM (2 of 2) [19/1/2003 1:35:39]

1.5.1 Clique

1.5.1 Clique

INPUT

OUTPUT

Input Description: A graph G=(V,E) . Problem: What is the largest S \subset V such that for all x,y \in S , (x,y) \in E ?

Implementations ● ● ●

DIMACS Implementation Challenges (FORTRAN) (rating 9) Neural-Networks for Cliques and Coloring (C) (rating 6) Combinatorica (Mathematica) (rating 1)

Related Problems ●

Independent Set

file:///E|/WEBSITE/FILES/CLIQUE05.HTM (1 of 2) [19/1/2003 1:35:40]

1.5.1 Clique ●

Vertex Cover

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/CLIQUE05.HTM (2 of 2) [19/1/2003 1:35:40]

1.6.2 Convex Hull

1.6.2 Convex Hull

INPUT

OUTPUT

Input Description: A set S of n points in d -dimensional space. Problem: Find the smallest convex polygon containing all the points of S .

Implementations ● ● ● ● ● ● ●

Qhull - higher dimensional convex hull program (C) (rating 10) LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 6) Clarkson's higher dimensional convex hull code (C) (rating 6) Joseph O'Rourke's Computational Geometry (C) (rating 6) Xtango and Polka Algorithm Animation Systems (C++) (rating 4) Moret and Shapiro's Algorithms P to NP (Pascal) (rating 3) Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 3)

file:///E|/WEBSITE/FILES2/CONVHULL.HTM (1 of 2) [19/1/2003 1:35:42]

1.6.2 Convex Hull

Related Problems ● ● ● ●

Simplifying Polygons Sorting Traveling Salesman Problem Voronoi Diagrams

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/CONVHULL.HTM (2 of 2) [19/1/2003 1:35:42]

Tools and Utilities

Tools and Utilities ● ● ● ●

Pascal to C translator Download g-zip sources from MIT gnu archive Guide to uncompressing ZIP files on Unix systems Unzip for Unix sources (tar format)

About the Book Send us Mail Go to Main Page

file:///E|/WEBSITE/TOOLS/TOOLS.HTM [19/1/2003 1:35:42]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (1 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (2 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (3 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (4 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (5 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (6 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (7 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (8 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (9 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (10 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (11 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (12 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (13 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (14 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (15 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (16 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (17 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (18 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (19 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (20 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (21 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (22 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (23 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (24 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (25 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (26 of 27) [19/1/2003 1:36:27]

Visual Links Index

file:///E|/WEBSITE/INFO/VIEW.HTM (27 of 27) [19/1/2003 1:36:27]

Interesting Data Files

Interesting Data Files

Below we point to several interesting data files. These files can be used as test data for several of the Implementation Challenges given in The Algorithm Design Manual , but are also likely to be of independent interest. ● ●

Airplane routes and schedules (from Roberto Tamassia -- local copy) Name-ethnicity data; over 3000 names from several nations (local copy)

About the Book Send us Mail Go to Main Page

file:///E|/WEBSITE/BIBLIO/TESTDATA/DATA.HTM [19/1/2003 1:36:29]

About the Ratings

About the Ratings Our implementations are classified by the problems they solve. For each problem, we rate all the implementations according to usefulness on a 1 to 10 scale, with 10 meaning the most useful and 1 least useful. The usefulness of an implementation is supposed to reflect how likely it is that someone looking for code will be happy with what they find. Our ratings are completely subjective and reflect our judgement. Please feel free to send corrections and feedback based on your experiences. Please note that many of these implementations with low usefulness ratings were produced primarily for expository purposes, and can serve as excellent models to study or emulate if not as ready-to-run code. Go to Main Page

file:///E|/WEBSITE/INFO/RATINGS.HTM [19/1/2003 1:36:29]

1.1 Data Structures

1.1 Data Structures Specific Algorithms ● ● ● ● ● ●

1.1.1 Dictionaries 1.1.2 Priority Queues 1.1.3 Suffix Trees and Arrays 1.1.4 Graph Data Structures 1.1.5 Set Data Structures 1.1.6 Kd-Trees

About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/MAJOR_SE/1_1.HTM [19/1/2003 1:36:29]

1.2 Numerical Problems

1.2 Numerical Problems Specific Algorithms ● ● ● ● ● ● ● ● ● ● ●

1.2.1 Solving Linear Equations 1.2.2 Bandwidth Reduction 1.2.3 Matrix Multiplication 1.2.4 Determinants and Permanents 1.2.5 Constrained and Unconstrained Optimization 1.2.6 Linear Programming 1.2.7 Random Number Generation 1.2.8 Factoring and Primality Testing 1.2.9 Arbitrary Precision Arithmetic 1.2.10 Knapsack Problem 1.2.11 Discrete Fourier Transform

About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/MAJOR_SE/1_2.HTM [19/1/2003 1:36:30]

1.3 Combinatorial Problems

1.3 Combinatorial Problems Specific Algorithms ● ● ● ● ● ● ● ● ● ●

1.3.1 Sorting 1.3.2 Searching 1.3.3 Median and Selection 1.3.4 Generating Permutations 1.3.5 Generating Subsets 1.3.6 Generating Partitions 1.3.7 Generating Graphs 1.3.8 Calendrical Calculations 1.3.9 Job Scheduling 1.3.10 Satisfiability

About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/MAJOR_SE/1_3.HTM [19/1/2003 1:36:31]

1.4 Graph Problems -- polynomial-time problems

1.4 Graph Problems -- polynomial-time problems Specific Algorithms ● ● ● ● ● ● ● ● ● ● ● ●

1.4.1 Connected Components 1.4.2 Topological Sorting 1.4.3 Minimum Spanning Tree 1.4.4 Shortest Path 1.4.5 Transitive Closure and Reduction 1.4.6 Matching 1.4.7 Eulerian Cycle / Chinese Postman 1.4.8 Edge and Vertex Connectivity 1.4.9 Network Flow 1.4.10 Drawing Graphs Nicely 1.4.11 Drawing Trees 1.4.12 Planarity Detection and Embedding

About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/MAJOR_SE/1_4.HTM [19/1/2003 1:36:31]

1.5 Graph Problems -- hard problems

1.5 Graph Problems -- hard problems Specific Algorithms ● ● ● ● ● ● ● ● ● ● ●

1.5.1 Clique 1.5.2 Independent Set 1.5.3 Vertex Cover 1.5.4 Traveling Salesman Problem 1.5.5 Hamiltonian Cycle 1.5.6 Graph Partition 1.5.7 Vertex Coloring 1.5.8 Edge Coloring 1.5.9 Graph Isomorphism 1.5.10 Steiner Tree 1.5.11 Feedback Edge/Vertex Set

About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/MAJOR_SE/1_5.HTM [19/1/2003 1:36:32]

1.6 Computational Geometry

1.6 Computational Geometry Specific Algorithms ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

1.6.1 Robust Geometric Primitives 1.6.2 Convex Hull 1.6.3 Triangulation 1.6.4 Voronoi Diagrams 1.6.5 Nearest Neighbor Search 1.6.6 Range Search 1.6.7 Point Location 1.6.8 Intersection Detection 1.6.9 Bin Packing 1.6.10 Medial-Axis Transformation 1.6.11 Polygon Partitioning 1.6.12 Simplifying Polygons 1.6.13 Shape Similarity 1.6.14 Motion Planning 1.6.15 Maintaining Line Arrangements 1.6.16 Minkowski Sum

About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/MAJOR_SE/1_6.HTM [19/1/2003 1:36:32]

1.7 Set and String Problems

1.7 Set and String Problems Specific Algorithms ● ● ● ● ● ● ● ● ●

1.7.1 Set Cover 1.7.2 Set Packing 1.7.3 String Matching 1.7.4 Approximate String Matching 1.7.5 Text Compression 1.7.6 Cryptography 1.7.7 Finite State Machine Minimization 1.7.8 Longest Common Substring 1.7.9 Shortest Common Superstring

About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/MAJOR_SE/1_7.HTM [19/1/2003 1:36:32]

C++ Language Implementations

Algorithm Implementations in C++ ● ● ● ● ● ● ● ● ● ● ●

LEDA - A Library of Efficient Data Types and Algorithms SimPack/Sim++ Simulation Toolkit Fire-Engine and Spare-Parts String and Language Algorithms Xtango and Polka Algorithm Animation Systems Algorithms in C++ -- Sedgewick Geolab -- Computational Geometry System Grail: finite automata and regular expressions Calendrical Calculations Stony Brook Project Implementations LINK -- Programming and Visualization Environment for Hypergraphs David Eppstein's Knuth-Morris-Pratt Algorithm and Minkowski sum code

About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/IMPLEMEN/CPP.HTM [19/1/2003 1:36:33]

C Language Implementations

Algorithm Implementations in C ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

Adaptive Simulated Annealing Joseph O'Rourke's Computational Geometry The Stanford GraphBase GraphViz -- graph layout programs Handbook of Algorithms and Data Structures Mike Trick's Graph Coloring Resources Joe Culberson's Graph Coloring Resources DIMACS Implementation Challenges Fire-Engine and Spare-Parts String and Language Algorithms Ranger - Nearest Neighbor Search in Higher Dimensions Xtango and Polka Algorithm Animation Systems Frank Ruskey's Combinatorial Generation Resources Goldberg's Network Optimization Codes Qhull - higher dimensional convex hull program Clarkson's higher dimensional convex hull code Fortune's 2D Voronoi diagram code Triangle: A Two-Dimensional Quality Mesh Generator Arrange - maintainance of arrangements with point location Linprog -- low dimensional linear programming LP_SOLVE: Linear Programming Code PARI - Package for Number Theory GraphEd -- Graph Editor and Layout Program Genocop -- Optimization via Genetic Algorithms TSP solvers FFTPACK -- Fourier Transform Library Neural-Networks for Cliques and Coloring PHYLIP -- inferring phylogenic trees Salowe's Rectilinear Steiner trees Skeletonization Software (2-D) SNNS - Stuttgart Neural Network Simulator agrep - Approximate General Regular Expression Pattern Matcher HT/DIG -- image compression codes CAP -- Contig Assembly Program Shape similarity testing via turning functions NAUTY -- Graph Isomorphism file:///E|/WEBSITE/IMPLEMEN/C.HTM (1 of 2) [19/1/2003 1:36:33]

C Language Implementations ● ●

POSIT - Propositional Satisfiability Testbed BIPM -- Bipartite Matching Codes

About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/IMPLEMEN/C.HTM (2 of 2) [19/1/2003 1:36:33]

Pascal Language Implementations

Algorithm Implementations in Pascal ● ● ● ● ●

Discrete Optimization Methods Handbook of Algorithms and Data Structures Moret and Shapiro's Algorithms P to NP Frank Ruskey's Combinatorial Generation Resources PHYLIP -- inferring phylogenic trees

About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/IMPLEMEN/PASCAL.HTM [19/1/2003 1:36:34]

FORTRAN Language Implementations

Algorithm Implementations in FORTRAN ● ● ● ● ● ●

Nijenhuis and Wilf: Combinatorial Algorithms DIMACS Implementation Challenges Netlib / TOMS -- Collected Algorithms of the ACM GEOMPACK - triangulation and convex decomposition codes FFTPACK -- Fourier Transform Library LAPACK and LINPACK -- Linear Algebra PACKages

About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/IMPLEMEN/FORTRAN.HTM [19/1/2003 1:36:34]

Mathematica Language Implementations

Algorithm Implementations in Mathematica ● ● ●

Combinatorica Mathematica -- Assorted Routines Neural-Networks for Cliques and Coloring

About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/IMPLEMEN/MATHEMAT.HTM [19/1/2003 1:36:34]

Lisp Language Implementations

Algorithm Implementations in Lisp ●

Calendrical Calculations

About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/IMPLEMEN/LISP.HTM [19/1/2003 1:36:35]

Algorithm Repository -- Most Wanted List

Most Wanted List Below are a list of problems for which we do not feel that we have identified adequate codes for the algorithm repository, and thus we would be particularly excited to hear of any programs you might know about. In many cases, the algorithms involved are not that difficult to implement, so this is also in invitation for students and faculty to select them for class projects and send us the resulting code. Our most wanted list includes: ●

●

●

●

●

● ●

Knapsack and Integer Partition -- We would like to see an implementation of the dynamic programming algorithm for the knapsack, with scaling. Associated with this would be an implementation of the primary bin-packing algorithms. Thinning or the Medial-Axis Transform -- Codes working on binary images and polygons are both wanted. I have already received more than one request for such a code. Chinese Postman Tour -- Such an algorithm should not be too complicated to implement, especially by using matching codes already available on this site. Graph Partition and Maxcut -- A flexible simulated annealing code for both problems would be greatly appreciated. Feedback arc/vertex sets -- It would be good to get an implementation of the (fairly simple) heuristics for feedback sets and see how well they do. Minkowski Sum -- I have already received more than one request for such a code. Set Cover -- I'd like to get more codes with heuristics for set cover and packing.

This algorithms repository is part of the ongoing algorithms-related research conducted here in the Department of Computer Science at the State University of New York at Stony Brook.

About the Book Send us Mail Go to Main Page

file:///E|/WEBSITE/INFO/WANTED.HTM [19/1/2003 1:36:35]

Algorithm Repository -- Citations

Citations Citations of the Stony Brook Algorithms Repository include: ● PC Webopaedia The algorithms repository is part of the ongoing algorithms-related research conducted here in the Department of Computer Science at the State University of New York at Stony Brook.

About the Book Send us Mail Go to Main Page

file:///E|/WEBSITE/INFO/CITATION.HTM [19/1/2003 1:36:36]

Practical Algorithm Design -- User Feedback

User Feedback This page allows users to comment on issues related to the Practical Algorithm Design website. Comments on the website, the upcoming book, implementations, and problems are all welcome, and will be posted on a separate page for access by other users of this site. To view comments left by other users, select: ●

User Comments

To leave a comment of your own, please fill out the following form: E-mail address:

Subject: Enter your comments below:

Reset

Submit Feedback

Return to the home page If you have problems with this page, please send E-mail to: [email protected] file:///E|/WEBSITE/FEEDBACK/FEEDBACK.HTM (1 of 2) [19/1/2003 1:36:36]

Practical Algorithm Design -- User Feedback

file:///E|/WEBSITE/FEEDBACK/FEEDBACK.HTM (2 of 2) [19/1/2003 1:36:36]

Goldberg's Network Optimization Codes

Goldberg's Network Optimization Codes The highest performance codes available for such network optimization problems as matching, shortest paths, and network flow have been developed by Andrew Goldberg and his collaborators. All are written in C. Their codes are available by ftp for non-commercial use, although a license is required for commercial use. For information on obtaining the codes, check out Andrew Goldberg's WWW page, http://www.neci.nj.nec.com/homepages/avg.html Their implementations of both Dijkstra and Bellman-Ford's algorithms for finding shortest paths in graphs is SPLIB, developed by Cherkassky, Goldberg, and Radzik. They report solving instances with over one million vertices in under two minutes on a Sun Sparc-10 workstation. Their code for finding a maximum cardinality bipartite matching of maximum weight shortest paths in graphs is CSA, developed by Goldberg and Kennedy. This code is based on a cost-scaling network flow algorithms. Their running times depend upon the density of the networks and weight distributions, but they report solving instances with over 30,000 vertices in a few minutes on a Sun Sparc-2 workstation. Their code for solving maximum-flow in graphs is PRF, developed by Cherkassky and Goldberg. They report solving instances with over 250,000 vertices in under two minutes on a Sun Sparc-10 workstation. For minimum-cost max-flow, the higher performance code available is CS, capable of solving instances of over 30,000 vertices in a few minutes on Sun Sparc-2 workstations.

●

Andrew Goldberg's home page

Problem Links ● ● ●

Network Flow (10) Matching (9) Shortest Path (9)

About the Book Send us Mail Go to Main Page This page last modified on Apr 23, 1997.

file:///E|/WEBSITE/IMPLEMEN/GOLDBERG/IMPLEMEN.HTM [19/1/2003 1:36:37]

LEDA - A Library of Efficient Data Types and Algorithms

LEDA - A Library of Efficient Data Types and Algorithms LEDA ("Library of Efficient Data types and Algorithms") has been developing since 1988 under the efforts of a group at the Max Planck Institut in Saarbrucken Germany (including Kurt Melhorn, Stefan Naher, Stefan Schirra, Christian Uhrig, and Christoph Burnikel). The success of LEDA has been a direct result of a continuous resource investment on the part of its algorithmically sophisticated development team. Implemented in C++ using templates, LEDA may be compiled on a wide range of systems (older compilers may not support templates, but most new C++ compilers provide facilities to implement this feature). The standard distribution contains source code, installation information, and a complete user's manual. Please note that LEDA is not in the public domain , but may be used freely for research and teaching. Commercial licenses are availabe through the LEDA home page. LEDA comprises an extensive collection of data structures and types. Libraries of algorithms using these data types are provided, with examples illustrating the ease with which algorithmic tasks may be accomplished given the LEDA data types.

● ● ●

Download LEDA Files (German site) Download LEDA files (local site) Go to LEDA Home Page

Problem Links ● ● ● ● ● ● ● ● ● ● ●

Dictionaries (10) Graph Data Structures (10) Priority Queues (9) Connected Components (8) Robust Geometric Primitives (8) Intersection Detection (7) Maintaining Line Arrangements (7) Matching (7) Planarity Detection and Embedding (7) Point Location (7) Range Search (7) file:///E|/WEBSITE/IMPLEMEN/LEDA/IMPLEMEN.HTM (1 of 2) [19/1/2003 1:36:38]

LEDA - A Library of Efficient Data Types and Algorithms ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

Searching (7) Shortest Path (7) Topological Sorting (7) Convex Hull (6) Minimum Spanning Tree (6) Nearest Neighbor Search (6) Transitive Closure and Reduction (6) Triangulation (6) Voronoi Diagrams (6) Generating Graphs (5) Arbitrary Precision Arithmetic (5) Network Flow (5) Set Data Structures (5) Graph Partition (4) Matrix Multiplication (4) Random Number Generation (4) Solving Linear Equations (3) Determinants and Permanents (2)

About the Book Send us Mail Go to Main Page This page last modified on Feb 23, 1996.

file:///E|/WEBSITE/IMPLEMEN/LEDA/IMPLEMEN.HTM (2 of 2) [19/1/2003 1:36:38]

Discrete Optimization Methods

Discrete Optimization Methods The Pascal procedures available in this archive are taken with permission from Discrete Optimization Algorithms with Pascal Programs by Maciej M. Syslo, Narsingh Deo, and Janusz S. Kowalik. This text was published in 1983 by Prentice-Hall, Inc., Englewood Cliffs, NJ. To our knowledge these programs are available nowhere else on the Internet. The field of discrete optimization (as viewed by the authors of the text above) consists of the areas of linear and integer programming, cover problems, knapsack problems, graph theory, network-flow problems, and scheduling. Their text covers these areas, using Pascal programs to elucidate methods of attacking discrete optimization problems. Those programs are downloadable from this site (see the previous page). Some notes on the programs themselves The methods used in the programs tend to speak for themselves, however for in-depth coverage of the problems and algorithms it is advised that a copy of the text be obtained. Many of the data types (particularly array data types) used in the Pascal procedures are assumed to be declared elsewhere (these are more "procedures" than complete programs), and are explicitly named only in the text. As a general rule, however, a naming convention is followed which should clear up most ambiguities. An array of integers which has indices ranging from 1 through N, which would be declared ARRAY[1..N] OF INTEGER , will be denoted by the data-type ARRN . Similarly, a two-dimensional array of integers which would be declared in Pascal as ARRAY[1..N, 1..M] OF INTEGER will be denoted by the data-type ARRNM in the procedures given.

● ● ●

Download files (local site) Files with driver programs and datafiles (local site) Index of files in this distribution

Problem Links ● ● ● ● ●

Set Cover (5) Set Packing (5) Shortest Path (5) Traveling Salesman Problem (5) Knapsack Problem (4) file:///E|/WEBSITE/IMPLEMEN/SYSLO/IMPLEMNT.HTM (1 of 2) [19/1/2003 1:36:39]

Discrete Optimization Methods ● ● ● ● ● ●

Network Flow (4) Job Scheduling (4) Vertex Coloring (4) Linear Programming (3) Matching (3) Minimum Spanning Tree (3)

About the Book Send us Mail Go to Main Page This page last modified on Apr 28, 1997.

file:///E|/WEBSITE/IMPLEMEN/SYSLO/IMPLEMNT.HTM (2 of 2) [19/1/2003 1:36:39]

Netlib / TOMS -- Collected Algorithms of the ACM

Netlib / TOMS -- Collected Algorithms of the ACM Netlib/TOMS -- Collected Algorithms of the ACM An early mechanism for the distribution of useful implementations of algorithms was CALGO , the Collected Algorithms of the ACM. It first appeared in Communications of the ACM in 1960, covering such famous algorithms as Floyd's linear-time build heap algorithm. More recently, it has been the province of ACM Transactions on Mathematical Software. Each algorithm/implementation is described in a brief journal article, with the implementation validated and collected. A total of 750 algorithms have appeared to date. Most of the codes are in Fortran, and of interest in numerical computing, although several interesting combinatorial algorithms have slithered into CALGO. Since the implementations have been refereed, presumably they are more reliable than most freely available software. All the implementations below are available by ftp or WWW from Netlib. Algorithms of combinatorial interest include: ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

Bandwidth -- 508 509 529 582 Convex Hull -- 523 Cryptography -- 536 Decision Trees -- 606 Fourier Transforms -- 545 Generating Partitions -- 403 Generating Subsets -- 515 Geometric Primitives -- 550 Hamiltonian Cycle -- 595 High Precision Arithmetic -- 693 Knapsack -- 632 Linear Programming -- 551 552 559 Matching -- 548 575 Matrix Inversion -- 645 Minimum Cut (facility location) -- 558 Minimum Spanning Trees -- 479 613 Priority Queues -- 561 Random Numbers -- 488 599 712 Scheduling -- 520 Shortest Path -- 562 Solving Linear Equations -- 533 576 578 Sorting -- 410 505 Sparse Matrix -- 601 Text Compression - 673 Traveling Salesman Problem -- 608 750

file:///E|/WEBSITE/IMPLEMEN/TOMS/IMPLEMEN.HTM (1 of 3) [19/1/2003 1:36:40]

Netlib / TOMS -- Collected Algorithms of the ACM ● ●

Triangulation -- 624 Unconstrained Optimization -- 500 566 611 630 702 734 739 744

A bibliography and keywords appear in Algorithm 620.

● ● ●

Collected Algorithms of the ACM Link to TOMS distribution Link to Netlib

Problem Links ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

Bandwidth Reduction (9) Constrained and Unconstrained Optimization (8) Solving Linear Equations (7) Discrete Fourier Transform (6) Generating Partitions (6) Hamiltonian Cycle (6) Knapsack Problem (6) Matrix Multiplication (6) Random Number Generation (6) Traveling Salesman Problem (6) Arbitrary Precision Arithmetic (5) Minimum Spanning Tree (5) Text Compression (5) Linear Programming (4) Job Scheduling (4) Shortest Path (4) Sorting (4) Triangulation (4) Convex Hull (3) Cryptography (3) Generating Subsets (3) Matching (3) Priority Queues (3) Voronoi Diagrams (3)

file:///E|/WEBSITE/IMPLEMEN/TOMS/IMPLEMEN.HTM (2 of 3) [19/1/2003 1:36:40]

Netlib / TOMS -- Collected Algorithms of the ACM

About the Book Send us Mail Go to Main Page This page last modified on May 17, 1997.

file:///E|/WEBSITE/IMPLEMEN/TOMS/IMPLEMEN.HTM (3 of 3) [19/1/2003 1:36:40]

Xtango and Polka Algorithm Animation Systems

Xtango and Polka Algorithm Animation Systems XTANGO is a general purpose algorithm animation system, developed by John Stasko of Georgia Tech, that supports programmers developing color, real-time, 2 & 1/2 dimensional, smooth animations of their own algorithms and programs. The focus of the system is on ease-of-use. XTANGO utilizes the pathtransition animation paradigm which helps move animation design to an abstract, high level. Programmers need not be graphics experts to develop their own animations. XTANGO is implemented on top of UNIX and the X11 Window System. It can be run on either color or black-and-white monitors. XTANGO is available for use via anonymous ftp from ftp.cc.gatech.edu in directory pub/people/stasko. Included with the XTANGO distribution is a large collection of animations. Several of which are quite fun to watch. My favorites include: ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

AVL trees (avl) Binomial heaps (binoheap) Boyer-Moore string matching (boyer) Bin Packing (bpack) Bubble sort (bsort) Convex Hull (conhull) Fast Fourier Transform (fft) Fibonacci heaps (fiboheap) Heapsort (heap) Knight's tour (knight) K-selection (ksel) Eight Queens (queens) Quicksort (quicksort) Radix sort (radix) Treaps (treap)

The basic process of animation consists of implementing the algorithm in C (another language can be used, but it must just produce a trace file which is read by a C program driver) and then deciding on the important events to be portrayed during the execution of the algorithm. These events then activate animation routines implemented in a separate file using the XTANGO animation package to create and manipulate objects (circles, squares, lines, and so on). Transitions on objects include movement, color change, resizing, and filling, as well as others. For example, the animation for binary search consists of a series of rectangles, each representaing one of the elements being searched. A bouncing circle hits the current dividing element, which changes color. The ball then bounces to the next dividing element and continues to do this until the desired element has been found. To learn more about XTANGO, see the September 1990 issue of IEEE Computer which has an article about the TANGO system, a ancestor of file:///E|/WEBSITE/IMPLEMEN/XTANGO/IMPLEMEN.HTM (1 of 3) [19/1/2003 1:36:41]

Xtango and Polka Algorithm Animation Systems

XTANGO. POLKA is a general purpose animation system that is particularly well-suited to building animations of programs, algorithms and computations, especially parallel computations. POLKA supports color, realtime, 2 & 1/2 dimensional, smooth animations. The focus of the system is on a balance of power and easeof-use. POLKA provides its own high-level abstractions to make the creation of animations easier and faster than with many other systems. Programmers need not be graphics experts to develop their own animations. POLKA also includes a hot new interactive front-end called SAMBA that can be used to generate animations from any type of program that can generate ASCII. POLKA is implemented in C++ on top of UNIX and the X11 Window System, and it requires the Motif or Athena widget set. (Because it supports Athena widgets, if you're running Linux, you should be able to get Polka to build there.) It can be run on either color (much better) or black-and-white monitors. POLKA is available for use via anonymous ftp from ftp.cc.gatech.edu under pub/people/stasko.

● ●

Link to Xtango distribution Download Files (local site)

Problem Links ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

Sorting (6) Dictionaries (5) Convex Hull (4) Median and Selection (4) Minimum Spanning Tree (4) Priority Queues (4) Shortest Path (4) Vertex Coloring (4) Bin Packing (3) Discrete Fourier Transform (3) Hamiltonian Cycle (3) Random Number Generation (3) String Matching (3) Traveling Salesman Problem (3) Connected Components (2) Matrix Multiplication (2) Topological Sorting (2) Finite State Machine Minimization (1) file:///E|/WEBSITE/IMPLEMEN/XTANGO/IMPLEMEN.HTM (2 of 3) [19/1/2003 1:36:41]

Xtango and Polka Algorithm Animation Systems ●

Intersection Detection (1)

About the Book Send us Mail Go to Main Page This page last modified on Mar 24, 1997.

file:///E|/WEBSITE/IMPLEMEN/XTANGO/IMPLEMEN.HTM (3 of 3) [19/1/2003 1:36:41]

Combinatorica

Combinatorica Combinatorica is a collection of over 230 algorithms for discrete mathematics and graph theory written in Mathematica. These routines have been designed to work together, enabling one to experiment with discrete structures and build prototype applications. Combinatorica has been widely used for both research and education. Although (in my totally unbiased opinion) Combinatorica is more comprehensive and better integrated than other libraries of combinatorial algorithms, it is also the slowest such system available. Credit for all of these properties is largely due to Mathematica, which provides a very high-level, functional, interpreted, and thus inefficient programming language. Combinatorica is best for finding quick solutions to small problems, and (if you can read Mathematica code) as a terse exposition of algorithms for translation into other languages. Combinatorica is best described in my book: Steven S. Skiena, Implementing Discrete Mathematics: Combinatorics and Graph Theory in Mathematica , Advanced Book Division, Addison-Wesley, Redwood City CA, June 1990. ISBN number 0-201-50943-1. Japanese translation published by Toppan, Tokyo, July 1992. Combinatorica is included with the standard Mathematica distribution in the directory Packages/DiscreteMath/Combinatorica.m . It can also be obtained by anonymous ftp from ftp.cs.sunysb.edu in the directory pub/Combinatorica. For this FTP site with the latest version of Combinatorica, databases of interesting graphs, and related programs, click here

● ● ●

Link to Combinatorica distribution Implementing Discrete Mathematics Download Files (local site)

Problem Links ● ● ● ● ● ●

Generating Graphs (8) Generating Partitions (7) Generating Permutations (7) Generating Subsets (7) Drawing Graphs Nicely (6) Drawing Trees (6) file:///E|/WEBSITE/IMPLEMEN/COMBINAT/IMPLEMEN.HTM (1 of 2) [19/1/2003 1:36:42]

Combinatorica ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

Graph Data Structures (6) Determinants and Permanents (4) Edge Coloring (4) Edge and Vertex Connectivity (4) Transitive Closure and Reduction (4) Vertex Cover (4) Connected Components (3) Eulerian Cycle / Chinese Postman (3) Graph Isomorphism (3) Hamiltonian Cycle (3) Matching (3) Minimum Spanning Tree (3) Network Flow (3) Shortest Path (3) Topological Sorting (3) Traveling Salesman Problem (3) Vertex Coloring (3) Longest Common Substring (2) Planarity Detection and Embedding (2) Sorting (2) Clique (1)

About the Book Send us Mail Go to Main Page This page last modified on Dec 20, 1996.

file:///E|/WEBSITE/IMPLEMEN/COMBINAT/IMPLEMEN.HTM (2 of 2) [19/1/2003 1:36:42]

file:///E|/WEBSITE/IMPLEMEN/GRAPHBAS/IMPLEMNT.HTM

The Stanford GraphBase The Stanford GraphBase is a collection of programs and datasets which generate and manipulate graphs and networks. This package is the work of Donald Knuth at Stanford University, and the most recent version of this software is always available by anonymous ftp from the Computer Science Department at Stanford (see previous page for link information). The programs themselves are written in CWEB , which is a mix of the C programming language and Knuth's TEX typesetting language. To install and use this package, therefore, it is necessary to first download and install CWEB on your system. We have made CWEB and the GraphBase available on this site, as well as providing links to the original sites. Files in GraphBase which have the .dat extension are data files, including dictionary-type data, map distance data, data for reconstructing the painting of the Mona Lisa, football score data, and so on. Much of the emphasis in the example GraphBase programs is on novel uses for graphs (for instance constructing word ladders: "flour - floor - flood - blood - brood - broad - bread"), while implementing efficient algorithmic methods to manipulate graphs and networks in general. The text The Stanford GraphBase: A Platform for Combinatorial Computing is available from AddisonWesley Publishing Company (ISBN 0-201-54275-7), and is a helpful overview of the system. This book shows the recreational approach of the author to the field of algorithms while providing a useful GraphBase reference.

● ● ● ●

Download GraphBase Files (Stanford site) Download CWEB Files (Stanford site) Download GraphBase and CWEB files (local site) Go to Stanford Computer Science Home Page

Problem Links ● ● ● ● ● ● ●

Generating Graphs (10) Graph Data Structures (6) Random Number Generation (6) Generating Partitions (5) Generating Permutations (5) Connected Components (4) Edge and Vertex Connectivity (4) file:///E|/WEBSITE/IMPLEMEN/GRAPHBAS/IMPLEMNT.HTM (1 of 2) [19/1/2003 1:36:43]

file:///E|/WEBSITE/IMPLEMEN/GRAPHBAS/IMPLEMNT.HTM ● ● ● ● ● ● ● ● ●

Feedback Edge/Vertex Set (4) Hamiltonian Cycle (4) Matching (4) Minimum Spanning Tree (4) Cryptography (3) Shortest Path (3) Topological Sorting (3) Voronoi Diagrams (3) Priority Queues (2)

About the Book Send us Mail Go to Main Page This page last modified on Feb 23, 1996.

file:///E|/WEBSITE/IMPLEMEN/GRAPHBAS/IMPLEMNT.HTM (2 of 2) [19/1/2003 1:36:43]

1.4.1 Connected Components

1.4.1 Connected Components

INPUT

OUTPUT

Input Description: A directed or undirected graph G . A start vertex . Problem: Traverse each edge and vertex of the connected component containing .

Implementations ● ● ● ● ● ● ●

LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 8) The Stanford GraphBase (C) (rating 4) GraphEd -- Graph Editor and Layout Program (C) (rating 4) Moret and Shapiro's Algorithms P to NP (Pascal) (rating 4) Combinatorica (Mathematica) (rating 3) Nijenhuis and Wilf: Combinatorial Algorithms (FORTRAN) (rating 2) Xtango and Polka Algorithm Animation Systems (C++) (rating 2)

file:///E|/WEBSITE/FILES/DFS_BFS4.HTM (1 of 2) [19/1/2003 1:36:43]

1.4.1 Connected Components

Related Problems ● ● ●

Edge and Vertex Connectivity Shortest Path Transitive Closure and Reduction

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/DFS_BFS4.HTM (2 of 2) [19/1/2003 1:36:43]

1.5.9 Graph Isomorphism

1.5.9 Graph Isomorphism

INPUT

OUTPUT

Input Description: Two graphs, g and h .} Problem: Find a (all) mappings f of the vertices of g to the vertices of h such that g and h are identical, ie. (x,y) is an edge of g iff (f(x),f(y)) is an edge of h .

Implementations ● ●

NAUTY -- Graph Isomorphism (C) (rating 10) Combinatorica (Mathematica) (rating 3)

Related Problems ● ● ●

Generating Graphs Shape Similarity Shortest Path

file:///E|/WEBSITE/FILES/GRAPHISM.HTM (1 of 2) [19/1/2003 1:36:44]

1.5.9 Graph Isomorphism ●

String Matching

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/GRAPHISM.HTM (2 of 2) [19/1/2003 1:36:44]

1.2.3 Matrix Multiplication

1.2.3 Matrix Multiplication

INPUT

OUTPUT

Input Description: An x x y matrix A , and an y x z matrix B . Problem: The x x z matrix A x B .

Implementations ● ● ● ● ●

() (rating 7) Mathematica -- Assorted Routines (Mathematica) (rating 6) Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 6) LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 4) Xtango and Polka Algorithm Animation Systems (C++) (rating 2)

Related Problems file:///E|/WEBSITE/FILES/MATRTION.HTM (1 of 2) [19/1/2003 1:36:44]

1.2.3 Matrix Multiplication ● ●

Solving Linear Equations Shortest Path

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/MATRTION.HTM (2 of 2) [19/1/2003 1:36:44]

1.6.14 Motion Planning

1.6.14 Motion Planning

INPUT

OUTPUT

Input Description: A polygonal-shaped robot in a given starting position in a room containing polygonal obstacles, with a desired ending position t . Problem: Find the shortest path in the room taking to t without going through any of the obstacles.

Implementations ●

Joseph O'Rourke's Computational Geometry (C) (rating 3)

Related Problems ● ● ●

Intersection Detection Minkowski Sum Shortest Path file:///E|/WEBSITE/FILES2/MOTINING.HTM (1 of 2) [19/1/2003 1:36:45]

1.6.14 Motion Planning

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/MOTINING.HTM (2 of 2) [19/1/2003 1:36:45]

1.4.9 Network Flow

1.4.9 Network Flow

INPUT

OUTPUT

Input Description: A graph G , where each edge (i,j) has a capacity c_{i,j} . A source node and sink node t . Problem: What is the maximum flow you can route from to t while respecting the capacity of each edge.

Implementations ● ● ● ● ● ● ● ●

Goldberg's Network Optimization Codes (C) (rating 10) DIMACS Implementation Challenges (FORTRAN) (rating 8) LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 5) Moret and Shapiro's Algorithms P to NP (Pascal) (rating 4) Discrete Optimization Methods (Pascal) (rating 4) Combinatorica (Mathematica) (rating 3) GraphEd -- Graph Editor and Layout Program (C) (rating 3) Nijenhuis and Wilf: Combinatorial Algorithms (FORTRAN) (rating 3) file:///E|/WEBSITE/FILES/NETWFLOW.HTM (1 of 2) [19/1/2003 1:36:45]

1.4.9 Network Flow

Related Problems ● ● ● ● ●

Edge and Vertex Connectivity Graph Partition Linear Programming Matching Shortest Path

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/NETWFLOW.HTM (2 of 2) [19/1/2003 1:36:45]

1.1.2 Priority Queues

1.1.2 Priority Queues

INPUT

OUTPUT

Input Description: A set of records with numerical or otherwise totally ordered keys. Problem: Build and maintain a data structures for quickly inserting and deleting records, while maintaining quick access to the smallest or largest key in the set.

Implementations ● ● ● ● ● ● ● ●

LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 9) Handbook of Algorithms and Data Structures (Pascal) (rating 7) SimPack/Sim++ Simulation Toolkit (C++) (rating 7) Xtango and Polka Algorithm Animation Systems (C++) (rating 4) Moret and Shapiro's Algorithms P to NP (Pascal) (rating 3) Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 3) The Stanford GraphBase (C) (rating 2) DIMACS Implementation Challenges (FORTRAN) (rating 1) file:///E|/WEBSITE/FILES/PRIOEUES.HTM (1 of 2) [19/1/2003 1:36:46]

1.1.2 Priority Queues

Related Problems ● ● ● ●

Dictionaries Median and Selection Shortest Path Sorting

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/PRIOEUES.HTM (2 of 2) [19/1/2003 1:36:46]

1.5.10 Steiner Tree

1.5.10 Steiner Tree

INPUT

Input Description: A graph G=(V,E) . A subset of vertices T \in V . Problem: Find the smallest tree connecting all the vertices of T .

Implementations ● ●

Salowe's Rectilinear Steiner trees (C) (rating 8) PHYLIP -- inferring phylogenic trees (Pascal) (rating 7)

Related Problems ● ●

Minimum Spanning Tree Shortest Path

file:///E|/WEBSITE/FILES/STEITREE.HTM (1 of 2) [19/1/2003 1:36:46]

OUTPUT

1.5.10 Steiner Tree

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/STEITREE.HTM (2 of 2) [19/1/2003 1:36:46]

1.4.5 Transitive Closure and Reduction

1.4.5 Transitive Closure and Reduction

INPUT

OUTPUT

Input Description: A directed graph G=(V,E) . Problem: For transitive closure, construct a graph G'=(V,E') with edge (i,j) \in E' iff there is a directed path from i to j in G . For transitive reduction, construct a small graph G'=(V,E') with a directed path from i to j in G' iff (i,j) \in E .

Implementations ● ●

LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 6) Combinatorica (Mathematica) (rating 4)

Related Problems ● ●

Connected Components Shortest Path

file:///E|/WEBSITE/FILES/TRANSURE.HTM (1 of 2) [19/1/2003 1:36:47]

1.4.5 Transitive Closure and Reduction

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/TRANSURE.HTM (2 of 2) [19/1/2003 1:36:47]

About the Book -- The Algorithm Design Manual

About the Book

Most professional programmers are not well equipped to tackle algorithm design problems. The Algorithm Design Manual, written by Steven S. Skiena and published by Telos/Springer-Verlag is uniquely designed to provide access to combinatorial algorithms technology for computer professionals and students. This book is considerably different than other books on algorithms. Why? ●

●

●

●

●

We reduce the design process to a sequence of questions to ask about the problem at hand. This provides a concrete path to take the non-expert from an initial problem statement to a reasonable solution. To provide a better perspective on how algorithm problems arise in the real world, we include a collection of `war stories', tales from our experience on real problems. The moral of these stories is that algorithm design and analysis is not just theory, but an important tool to be pulled out and used as needed. Since the practical person is usually looking for a program more than an algorithm, we provide pointers to solid implementations whenever they are available. We have collected these implementations on an enclosed CD-ROM and at the book WWW site, http://www.cs.sunysb.edu/~algorith for easy retrieval. With these implementations available, the critical aspect in algorithm design becomes properly modeling your application, instead of becoming intimate with the details of the actual algorithm. This focus permeates the entire book. Since finding out what is known about an algorithmic problem can be a difficult task, we provide a catalog of the 75 most important algorithmic problems as a major component of this book. By browsing through this catalog, the reader can quickly identify what their problem called, what is known about it, and how they should proceed to solve it. As an aid in problem identification, we include a pair of `before' and `after' pictures for each problem, illustrating the required input and output specifications. The algorithm catalog spans numerical problems and data structures as well as graph, string, and geometric algorithms. For each problem in the catalog, we provide an honest and convincing motivation showing how it arises in applications. If we could not find such an application, then the

file:///E|/WEBSITE/INFO/FORM.HTM (1 of 3) [19/1/2003 1:36:48]

About the Book -- The Algorithm Design Manual

problem doesn't appear in this book. Equally important is what we do not do in this book. We do not stress the mathematical analysis of algorithms, leaving most of the analysis as informal arguments. You will not find a single theorem anywhere in this book. But what is a manual without software? This book comes with a substantial electronic supplement, a ISO9660 compatible, multiplatform CD-ROM, which can be viewed using Netscape, Microsoft Explorer, or any other WWW browser. This CD-ROM contains: ●

●

●

A complete hypertext version of the full printed book. Indeed, the extensive cross-references within the text are best followed using the hypertext version. The source code and URLs for all cited implementations, mirroring the Algorithm Repository WWW site. Programs in C, C++, Fortran, and Pascal are included, providing an average of four different implementations for each algorithmic problem. Over 30 hours of audio lectures on the design and analysis of algorithms are provided, all keyed to on-line lecture notes. Following these lectures provides another approach to learning algorithm design techniques.

Together, this book covers material sufficient for a standard Introduction to Algorithms course. Its assumes the reader has completed the equivalent of a second programming course, typically titled Data Structures or Computer Science II . Special textbook oriented-features include: ●

●

●

●

In addition to standard pen-and-paper exercises, this book includes ``implementation challenges'' suitable for teams or individual students. These projects and the applied focus of the text can be used to provide a new laboratory focus to the traditional algorithms course. ``Take-home lessons'' at the beginning of each chapter emphasize the concepts to be gained from the chapter. This book stresses design over analysis. It is suitable for both traditional lecture courses, and the new ``active learning'' method, where the professor does not lecture instead guides student groups to solve real problems. The ``war stories'' provide a great introduction to the active learning method. A full set of lecture slides for teaching this course is available on the CD-ROM, keyed to unique on-line audio lectures covering a full semester algorithm course.

``I have no doubt that it will become a classic the day it is published. It has all the right ingredients: rich contents, friendly, personal language, subtle humor, the right references, and a plethora of pointers to resources.'' -- P. Takis Metaxas, Wellesley College. ``A major theme that runs through the book is that the most important technique to solve an algorithmic problem from the real world is to learn how to model the problem well. I did not believe this before; the

file:///E|/WEBSITE/INFO/FORM.HTM (2 of 3) [19/1/2003 1:36:48]

About the Book -- The Algorithm Design Manual

book did an admirable job of convincing me that there is considerable truth in it.'' -- Giri Narasimhan, The University of Memphis. ``The questions on problem solving are good enough that they ought to be talked about in every programming class in the undergraduate curriculum.'' -- Ron Danielson, Santa Clara University. Check out the preface and table of contents for more information. You may order this book, and are encouraged to do so. You might also be interested in my previous book, Implementing Discrete Mathematics . Please leave your name and address to receive additional information about the book and notification of significant upgrades to this site when they occur.

If your WWW client does not support forms, please send an e-mail to [email protected] with your name, e-mail address, and mailing address for further information. First Name: Last Name : E-mail : Company Name : Number and Street : City : State : Zip : Phone : Reset

Register

Return to the home page If you have problems with this page, please send E-mail to: [email protected] file:///E|/WEBSITE/INFO/FORM.HTM (3 of 3) [19/1/2003 1:36:48]

Adaptive Simulated Annealing

Adaptive Simulated Annealing Adaptive Simulated Annealing (ASA) is a C language package which provides the framework and mechanisms for optimization of complex systems using simulated annealing. Users may incorporate these routines into existing C programs, or build applications from the sample ASA applications. Most parameters are user-definable -- allowing for highly customizable optimization algorithms. The code is maintained by Lester Ingber .

● ● ●

Download ASA files (ingber.com) Link to Lester Ingber's ASA page Download Files (local site)

Problem Links ●

Constrained and Unconstrained Optimization (6)

About the Book Send us Mail Go to Main Page This page last modified on Mar 11, 1996.

file:///E|/WEBSITE/IMPLEMEN/ASA/IMPLEMEN.HTM [19/1/2003 1:36:48]

Genocop -- Optimization via Genetic Algorithms

Genocop -- Optimization via Genetic Algorithms Genocop, by Zbigniew Michalewicz, is a genetic algorithm-based program for constrained and unconstrained optimization, written in C. The Genocop system aims at finding a global optimum (minimum or maximum: this is one of the input parameters) of a function; additional linear constraints (equations and inequalities) can be specified as well. The current version of Genocop should run without changes on any BSD-UN*X system (preferably on a Sun SPARC machine). This program can also be run on a DOS system. This software is copyright by Zbigniew Michalewicz. Permission is granted to copy and use the software for scientific, noncommercial purposes only. The software is provided "as is", i.e., without any warranties.

● ●

Original FTP site Download Files (local site)

Problem Links ●

Constrained and Unconstrained Optimization (5)

About the Book Send us Mail Go to Main Page This page last modified on Aug 9, 1996.

file:///E|/WEBSITE/IMPLEMEN/GENOCOP/IMPLEMEN.HTM [19/1/2003 1:36:49]

1.2.6 Linear Programming

1.2.6 Linear Programming

INPUT

OUTPUT

Input Description: A set of linear inequalities, a linear objective function. Problem: Find the assignment to the variables maximizing the objective function while satisfying all inequalities.

Implementations ● ● ● ● ●

LP_SOLVE: Linear Programming Code (C) (rating 9) Linprog -- low dimensional linear programming (C) (rating 4) Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 4) Discrete Optimization Methods (Pascal) (rating 3) Algorithms in C++ -- Sedgewick (C++) (rating 2)

Related Problems file:///E|/WEBSITE/FILES/LINEMING.HTM (1 of 2) [19/1/2003 1:36:50]

1.2.6 Linear Programming

● ● ●

Knapsack Problem Network Flow Constrained and Unconstrained Optimization

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/LINEMING.HTM (2 of 2) [19/1/2003 1:36:50]

1.2.7 Random Number Generation

1.2.7 Random Number Generation

INPUT

OUTPUT

Input Description: Nothing, or perhaps a seed. Problem: Generate a sequence of random integers.

Implementations ● ● ● ● ● ●

SimPack/Sim++ Simulation Toolkit (C++) (rating 7) The Stanford GraphBase (C) (rating 6) Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 6) LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 4) Algorithms in C++ -- Sedgewick (C++) (rating 3) Xtango and Polka Algorithm Animation Systems (C++) (rating 3)

Related Problems file:///E|/WEBSITE/FILES/RANDBERS.HTM (1 of 2) [19/1/2003 1:36:50]

1.2.7 Random Number Generation

● ● ● ●

Generating Partitions Generating Permutations Generating Subsets Constrained and Unconstrained Optimization

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/RANDBERS.HTM (2 of 2) [19/1/2003 1:36:50]

1.3.10 Satisfiability

1.3.10 Satisfiability

INPUT

OUTPUT

Input Description: A set of clauses in conjunctive normal form. Problem: Is there a truth assignment to the boolean variables such that every clause is satisfied?

Implementations ● ●

DIMACS Implementation Challenges (FORTRAN) (rating 8) POSIT - Propositional Satisfiability Testbed (C) (rating 8)

Related Problems ● ●

Finite State Machine Minimization Traveling Salesman Problem

file:///E|/WEBSITE/FILES/SATILITY.HTM (1 of 2) [19/1/2003 1:36:51]

1.3.10 Satisfiability ●

Constrained and Unconstrained Optimization

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/SATILITY.HTM (2 of 2) [19/1/2003 1:36:51]

Fortune's 2D Voronoi diagram code

Fortune's 2D Voronoi diagram code This is a fairly widely-used 2D code for Voronoi diagrams and Delauney triangulations, written in C by Steve Fortune of Bell Laboratories. It is based on Fortune's sweepline algorithm for Voronoi diagrams, and is likely to be the right code to try first. ● ● ●

Netlib Voronoi source directory Steve Fortune's home page Download Files (local site)

Problem Links ● ●

Voronoi Diagrams (9) Triangulation (7)

About the Book Send us Mail Go to Main Page This page last modified on Apr 23, 1996.

file:///E|/WEBSITE/IMPLEMEN/FORTUNE/IMPLEMEN.HTM [19/1/2003 1:36:51]

Qhull - higher dimensional convex hull program

Qhull - higher dimensional convex hull program Qhull is a general dimension code for computing convex hulls, Delaunay triangulations, Voronoi vertices, furthest-site Voronoi vertices, and halfspace intersections. It appears to be the choice for higherdimensional convex hull applications. Qhull is written in C, and implements the divide-and-conquer Quickhull algorithm. It is fast and roundoff errors from floating point arithmetic. Qhull has been widely used in scientific applications and has a well-maintained ●

● ● ●

home page . Qhull home page Latest news Download Files (local site)

Problem Links ● ● ● ●

Convex Hull (10) Voronoi Diagrams (7) Triangulation (6) Intersection Detection (5)

About the Book Send us Mail Go to Main Page This page last modified on Apr 8, 1997.

file:///E|/WEBSITE/IMPLEMEN/QHULL/IMPLEMEN.HTM [19/1/2003 1:36:52]

Joseph O'Rourke's Computational Geometry

Joseph O'Rourke's Computational Geometry Joseph O'Rourke is Olin Professor of Computer Science at Smith College in Northampton, Massachusetts. His text Computational Geometry in C has become one of the definitive computational geometry resources. The programs coded in the text have been made freely available by anonymous ftp from Smith College and have been included at this site as well. In this distribution are standard C language routines for simple computational geometric methods (determining whether a point lies inside a polygon, for instance) as well as robust implementations of complex computational geometry algorithms. Addressed are problems in motion planning, nearest neighbor determination (through the use of Delaunay triangulations and Voronoi diagrams), polygon intersection, convex hull computation, and polygon triangulation.

● ● ●

Download Files (Smith College) Download Files (local site) Go to Joseph O'Rourke 's Home Page

Problem Links ● ● ● ● ● ● ● ●

Convex Hull (6) Robust Geometric Primitives (6) Intersection Detection (5) Nearest Neighbor Search (5) Voronoi Diagrams (4) Motion Planning (3) Point Location (3) Triangulation (3)

About the Book Send us Mail Go to Main Page This page last modified on Feb 23, 1996.

file:///E|/WEBSITE/IMPLEMEN/OROURKE/IMPLEMEN.HTM [19/1/2003 1:36:53]

1.6.5 Nearest Neighbor Search

1.6.5 Nearest Neighbor Search

INPUT

OUTPUT

Input Description: A set S of n points in d dimensions; a query point q . Problem: Which point in S is closest to q ?

Implementations ● ● ● ●

Ranger - Nearest Neighbor Search in Higher Dimensions (C) (rating 7) LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 6) Joseph O'Rourke's Computational Geometry (C) (rating 5) Moret and Shapiro's Algorithms P to NP (Pascal) (rating 3)

Related Problems file:///E|/WEBSITE/FILES2/NEARHBOR.HTM (1 of 2) [19/1/2003 1:36:53]

1.6.5 Nearest Neighbor Search ● ● ● ●

Kd-Trees Point Location Range Search Voronoi Diagrams

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/NEARHBOR.HTM (2 of 2) [19/1/2003 1:36:53]

1.6.7 Point Location

1.6.7 Point Location

INPUT

OUTPUT

Input Description: A decomposition of the plane into polygonal regions, and a query point q . Problem: Which region contains the query point q ?

Implementations ● ● ● ●

LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 7) Arrange - maintainance of arrangements with point location (C) (rating 6) Moret and Shapiro's Algorithms P to NP (Pascal) (rating 3) Joseph O'Rourke's Computational Geometry (C) (rating 3)

Related Problems file:///E|/WEBSITE/FILES2/POINTION.HTM (1 of 2) [19/1/2003 1:36:54]

1.6.7 Point Location ● ● ● ● ●

Kd-Trees Maintaining Line Arrangements Nearest Neighbor Search Range Search Voronoi Diagrams

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/POINTION.HTM (2 of 2) [19/1/2003 1:36:54]

1.6.10 Medial-Axis Transformation

1.6.10 Medial-Axis Transformation

INPUT

OUTPUT

Input Description: A polygon or polyhedron P . Problem: What is the set of points within P which have more than one closest point on the boundary of P?

Implementations ●

Skeletonization Software (2-D) (C) (rating 9)

Related Problems ● ● ●

Minkowski Sum Shape Similarity Voronoi Diagrams file:///E|/WEBSITE/FILES2/00000011.HTM (1 of 2) [19/1/2003 1:36:54]

1.6.10 Medial-Axis Transformation

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/00000011.HTM (2 of 2) [19/1/2003 1:36:54]

1.6.3 Triangulation

1.6.3 Triangulation

INPUT

OUTPUT

Input Description: A set of points, or a polyhedron Problem: Partition the interior of the point set or polyhedron into triangles.

Implementations ● ● ● ● ● ● ● ●

Triangle: A Two-Dimensional Quality Mesh Generator (C) (rating 9) GEOMPACK - triangulation and convex decomposition codes (FORTRAN) (rating 8) Fortune's 2D Voronoi diagram code (C) (rating 7) LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 6) Qhull - higher dimensional convex hull program (C) (rating 6) GraphEd -- Graph Editor and Layout Program (C) (rating 4) Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 4) Joseph O'Rourke's Computational Geometry (C) (rating 3)

file:///E|/WEBSITE/FILES2/TRIAIONS.HTM (1 of 2) [19/1/2003 1:36:55]

1.6.3 Triangulation

Related Problems ● ●

Polygon Partitioning Voronoi Diagrams

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/TRIAIONS.HTM (2 of 2) [19/1/2003 1:36:55]

Nijenhuis and Wilf: Combinatorial Algorithms

Nijenhuis and Wilf: Combinatorial Algorithms Nijenhuis and Wilf's Combinatorial Algorithms , published by Academic Press in 1978, specializes in algorithms for constructing basic combinatorial objects such as permutations, subsets, and partitions; both randomly and sequentially. Such algorithms are often very short but hard to locate and usually are surprisingly subtle. Fortran programs for all of the algorithms are provided, as well as a discussion of the theory behind each of them. The programs are usually short enough that it is reasonable to translate directly into a more modern programming language, as I did with many of them in writing Combinatorica . Descriptions of more recent algorithms for several problems, without code, are provided in Wilf's Combinatorial Algorithms, an update , published by SIAM in 1989. These programs are now available here on our algorithm repository WWW site. We tracked them down from Neil Sloane, who had them on a magnetic tape where the authors did not! In their book, Nijenhuis and Wilf set the proper standard of statistically testing the output distribution of each of the random generators to establish that they really appear uniform. We encourage you to do the same before using these programs to verify that nothing has been lost in transit.

● ● ●

Link to Wilf's Home Page -- many interesting things Download Files (local site) Files with driver programs and test data (local site)

Problem Links ● ● ● ● ● ● ● ● ● ● ● ●

Generating Partitions (8) Generating Permutations (8) Generating Subsets (8) Hamiltonian Cycle (5) Determinants and Permanents (4) Generating Graphs (4) Eulerian Cycle / Chinese Postman (3) Minimum Spanning Tree (3) Network Flow (3) Sorting (3) Vertex Coloring (3) Connected Components (2)

file:///E|/WEBSITE/IMPLEMEN/WILF/IMPLEMEN.HTM (1 of 2) [19/1/2003 1:36:55]

Nijenhuis and Wilf: Combinatorial Algorithms

About the Book Send us Mail Go to Main Page This page last modified on Apr 28, 1997.

file:///E|/WEBSITE/IMPLEMEN/WILF/IMPLEMEN.HTM (2 of 2) [19/1/2003 1:36:55]

1.5.5 Hamiltonian Cycle

1.5.5 Hamiltonian Cycle

INPUT

OUTPUT

Input Description: A graph G = (V,E) . Problem: Find an ordering of the vertices such that each vertex is visited exactly once.

Implementations ● ● ● ● ●

Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 6) Nijenhuis and Wilf: Combinatorial Algorithms (FORTRAN) (rating 5) The Stanford GraphBase (C) (rating 4) Combinatorica (Mathematica) (rating 3) Xtango and Polka Algorithm Animation Systems (C++) (rating 3)

Related Problems file:///E|/WEBSITE/FILES/HAMIYCLE.HTM (1 of 2) [19/1/2003 1:36:56]

1.5.5 Hamiltonian Cycle ● ●

Eulerian Cycle / Chinese Postman Traveling Salesman Problem

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/HAMIYCLE.HTM (2 of 2) [19/1/2003 1:36:56]

1.4.6 Matching

1.4.6 Matching

INPUT

OUTPUT

Input Description: A (weighted) graph G=(V,E) . Problem: Find the largest size set of edges S \in E such that each vertex in V is incident to at most one edge of S .

Implementations ● ● ● ● ● ● ● ●

DIMACS Implementation Challenges (FORTRAN) (rating 9) Goldberg's Network Optimization Codes (C) (rating 9) BIPM -- Bipartite Matching Codes (C) (rating 8) LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 7) The Stanford GraphBase (C) (rating 4) Moret and Shapiro's Algorithms P to NP (Pascal) (rating 4) Combinatorica (Mathematica) (rating 3) Discrete Optimization Methods (Pascal) (rating 3) file:///E|/WEBSITE/FILES/00000034.HTM (1 of 2) [19/1/2003 1:36:57]

1.4.6 Matching ● ●

Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 3) Algorithms in C++ -- Sedgewick (C++) (rating 2)

Related Problems ● ● ● ● ●

Determinants and Permanents Eulerian Cycle / Chinese Postman Network Flow Job Scheduling Set Cover

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/00000034.HTM (2 of 2) [19/1/2003 1:36:57]

A compendium of NP optimization problems

A compendium of NP optimization problems Pierluigi Crescenzi , [email protected] and Viggo Kann , [email protected]

This is a preliminary version (April 1997) of the catalog of NP optimization problems. Please send any comment or suggestion to one of the two authors. A printed version of the compendium will appear in: Ausiello, G., Crescenzi, P., Gambosi, G., Kann, V., Marchetti Spaccamela, A., and Protasi, M., Approximate solution of NP-hard optimization problems. Springer-Verlag, 1997/1998 The latest version of the compendium is available on WWW at http://www.nada.kth.se/theory/compendium/ . There you will also find WWW forms to report new problems, new results on existing problems and errors.

Abstract: Due to the fact that no NP-complete problem can be solved in polynomial time (unless P=NP), many approximability results (both positive and negative) of NP-hard optimization problems have appeared in the technical literature. In this compendium, we collect a large number of these results.

●

●

Introduction ❍ NPO Problems: Definitions and Preliminaries ❍ Approximate Algorithms and Approximation Classes ❍ Completeness in Approximation Classes ❍ A list of NPO problems ❍ Improving the compendium Graph Theory ❍ Covering and Partitioning ❍ Subgraphs and Supergraphs ❍ Vertex Ordering

file:///E|/COMPEND/COMPED19/COMPENDI.HTM (1 of 2) [19/1/2003 1:36:58]

A compendium of NP optimization problems

Iso- and Other Morphisms ❍ Miscellaneous Network Design ❍ Spanning Trees ❍ Cuts and Connectivity ❍ Routing Problems ❍ Flow Problems ❍ Miscellaneous Sets and Partitions ❍ Covering, Hitting, and Splitting ❍ Weighted Set Problems Storage and Retrieval ❍ Data Storage ❍ Compression and Representation ❍ Miscellaneous Sequencing and Scheduling ❍ Sequencing on One Processor ❍ Multiprocessor Scheduling ❍ Shop Scheduling ❍ Miscellaneous Mathematical Programming Algebra and Number Theory Games and Puzzles Logic Program Optimization Miscellaneous References Index About this document ... ❍

●

●

●

●

● ● ● ● ● ● ● ● ●

Viggo Kann Mon Apr 21 13:07:14 MET DST 1997

file:///E|/COMPEND/COMPED19/COMPENDI.HTM (2 of 2) [19/1/2003 1:36:58]

1.6.9 Bin Packing

1.6.9 Bin Packing

INPUT

OUTPUT

Input Description: A set of n items with sizes d_1,...,d_n . A set of m bins with capacity c_1,...,c_m . Problem: How do you store the set of items using the fewest number of bins?

Implementations ●

Xtango and Polka Algorithm Animation Systems (C++) (rating 3)

Related Problems ● ● ●

Knapsack Problem Job Scheduling Set Packing

file:///E|/WEBSITE/FILES2/BIN_KING.HTM (1 of 2) [19/1/2003 1:36:58]

1.6.9 Bin Packing

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/BIN_KING.HTM (2 of 2) [19/1/2003 1:36:58]

1.6.12 Simplifying Polygons

1.6.12 Simplifying Polygons

INPUT

OUTPUT

Input Description: A polygon or polyhedron p , with n vertices. Problem: Find a polygon or polyhedron p' with n' vertices, where the shape of p' is close to p while n' << n .

Implementations ●

Skeletonization Software (2-D) (C) (rating 5)

Related Problems ● ● ●

Convex Hull Discrete Fourier Transform Minkowski Sum file:///E|/WEBSITE/FILES2/SIMPGONS.HTM (1 of 2) [19/1/2003 1:36:59]

1.6.12 Simplifying Polygons

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/SIMPGONS.HTM (2 of 2) [19/1/2003 1:36:59]

1.6.13 Shape Similarity

1.6.13 Shape Similarity

INPUT

Input Description: Two polygonal shapes, P_1 and P_2 . Problem: How similar are P_1 and P_2 ?

Implementations ● ●

SNNS - Stuttgart Neural Network Simulator (C) (rating 7) Shape similarity testing via turning functions (C) (rating 6)

Related Problems ● ●

Graph Isomorphism Medial-Axis Transformation

file:///E|/WEBSITE/FILES2/SHAPRITY.HTM (1 of 2) [19/1/2003 1:36:59]

OUTPUT

1.6.13 Shape Similarity

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/SHAPRITY.HTM (2 of 2) [19/1/2003 1:36:59]

1.6.11 Polygon Partitioning

1.6.11 Polygon Partitioning

INPUT

OUTPUT

Input Description: A polygon or polyhedron P . Problem: How can P be partitioned into a small number of simple (typically convex) pieces?

Implementations ●

GEOMPACK - triangulation and convex decomposition codes (FORTRAN) (rating 8)

Related Problems ● ●

Set Cover Triangulation

file:///E|/WEBSITE/FILES2/POLONING.HTM (1 of 2) [19/1/2003 1:37:00]

1.6.11 Polygon Partitioning

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/POLONING.HTM (2 of 2) [19/1/2003 1:37:00]

1.4.3 Minimum Spanning Tree

1.4.3 Minimum Spanning Tree

INPUT

OUTPUT

Input Description: A graph G = (V,E) with weighted edges. Problem: The subset of E of G of minimum weight which forms a tree on V .

Implementations ● ● ● ● ● ● ● ● ●

LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 6) Moret and Shapiro's Algorithms P to NP (Pascal) (rating 5) Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 5) The Stanford GraphBase (C) (rating 4) Xtango and Polka Algorithm Animation Systems (C++) (rating 4) Combinatorica (Mathematica) (rating 3) Algorithms in C++ -- Sedgewick (C++) (rating 3) Discrete Optimization Methods (Pascal) (rating 3) Nijenhuis and Wilf: Combinatorial Algorithms (FORTRAN) (rating 3) file:///E|/WEBSITE/FILES/MINITREE.HTM (1 of 2) [19/1/2003 1:37:00]

1.4.3 Minimum Spanning Tree

Related Problems ● ● ●

Set Data Structures Steiner Tree Traveling Salesman Problem

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/MINITREE.HTM (2 of 2) [19/1/2003 1:37:00]

1.6.15 Maintaining Line Arrangements

1.6.15 Maintaining Line Arrangements

INPUT

OUTPUT

Input Description: A set of lines and line segments l_1,...,\l_n . Problem: What is the decomposition of the plane defined by l_1,...,\l_n ?

Implementations ● ●

Arrange - maintainance of arrangements with point location (C) (rating 9) LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 7)

Related Problems ● ●

Robust Geometric Primitives Intersection Detection

file:///E|/WEBSITE/FILES2/MAINENTS.HTM (1 of 2) [19/1/2003 1:37:01]

1.6.15 Maintaining Line Arrangements ●

Point Location

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/MAINENTS.HTM (2 of 2) [19/1/2003 1:37:01]

1.3.7 Generating Graphs

1.3.7 Generating Graphs

INPUT

OUTPUT

Input Description: Parameters describing the desired graph, such as the number of vertices n , the number of edges m , or the edge probability p . Problem: Generate (1) all, or (2) a random, or (3) the next graph satisfying the parameters.

Implementations ● ● ● ● ●

The Stanford GraphBase (C) (rating 10) Combinatorica (Mathematica) (rating 8) Frank Ruskey's Combinatorial Generation Resources (Pascal) (rating 7) LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 5) Nijenhuis and Wilf: Combinatorial Algorithms (FORTRAN) (rating 4)

Related Problems file:///E|/WEBSITE/FILES/GENRAPHS.HTM (1 of 2) [19/1/2003 1:37:01]

1.3.7 Generating Graphs

● ●

Generating Permutations Graph Isomorphism

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/GENRAPHS.HTM (2 of 2) [19/1/2003 1:37:01]

1.2.11 Discrete Fourier Transform

1.2.11 Discrete Fourier Transform

INPUT

OUTPUT

Input Description: A sequence of n real or complex values h_i , 0 \leq i \leq n-1 , sampled at uniform intervals from a function h . Problem: The discrete Fourier transform H of h , H_m = \sum_{k=0}^{n-1} h_k e^{2 \pi i k m / n} , 0 \leq m \leq n-1 .

Implementations ● ● ● ● ●

FFTPACK -- Fourier Transform Library (C) (rating 10) Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 6) Moret and Shapiro's Algorithms P to NP (Pascal) (rating 4) Xtango and Polka Algorithm Animation Systems (C++) (rating 3) Algorithms in C++ -- Sedgewick (C++) (rating 2)

file:///E|/WEBSITE/FILES/FOUSFORM.HTM (1 of 2) [19/1/2003 1:37:02]

1.2.11 Discrete Fourier Transform

Related Problems ● ● ●

Arbitrary Precision Arithmetic Simplifying Polygons Text Compression

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/FOUSFORM.HTM (2 of 2) [19/1/2003 1:37:02]

1.5.8 Edge Coloring

1.5.8 Edge Coloring

INPUT

OUTPUT

Input Description: A graph G=(V,E) . Problem: What is the smallest set of colors needed to color the edges of E such that no two edges with the same color share a vertex in common?

Implementations ● ● ● ●

Stony Brook Project Implementations (C++) (rating 6) Combinatorica (Mathematica) (rating 4) Joe Culberson's Graph Coloring Resources (C) (rating 4) Mike Trick's Graph Coloring Resources (C) (rating 4)

Related Problems file:///E|/WEBSITE/FILES/EDGERING.HTM (1 of 2) [19/1/2003 1:37:02]

1.5.8 Edge Coloring ● ●

Job Scheduling Vertex Coloring

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/EDGERING.HTM (2 of 2) [19/1/2003 1:37:02]

1.3.3 Median and Selection

1.3.3 Median and Selection

INPUT

OUTPUT

Input Description: A set of n numbers or keys. Problem: Find the item which is smaller than half of the items and bigger than half the items.

Implementations ● ●

Handbook of Algorithms and Data Structures (Pascal) (rating 6) Xtango and Polka Algorithm Animation Systems (C++) (rating 4)

Related Problems ● ●

Priority Queues Sorting

file:///E|/WEBSITE/FILES/MEDIAN03.HTM (1 of 2) [19/1/2003 1:37:03]

1.3.3 Median and Selection

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/MEDIAN03.HTM (2 of 2) [19/1/2003 1:37:03]

1.3.1 Sorting

1.3.1 Sorting

INPUT

OUTPUT

Input Description: A set of n items. Problem: Arrange the items in increasing order.

Implementations ● ● ● ● ● ● ●

Handbook of Algorithms and Data Structures (Pascal) (rating 7) Moret and Shapiro's Algorithms P to NP (Pascal) (rating 7) Xtango and Polka Algorithm Animation Systems (C++) (rating 6) Algorithms in C++ -- Sedgewick (C++) (rating 5) Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 4) Nijenhuis and Wilf: Combinatorial Algorithms (FORTRAN) (rating 3) Combinatorica (Mathematica) (rating 2)

file:///E|/WEBSITE/FILES/SORTING2.HTM (1 of 2) [19/1/2003 1:37:03]

1.3.1 Sorting

Related Problems ● ● ● ● ● ●

Convex Hull Dictionaries Median and Selection Priority Queues Searching Topological Sorting

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/SORTING2.HTM (2 of 2) [19/1/2003 1:37:03]

Plugins for use with the CDROM

Plug-Ins Available On This CD-ROM: One of these plug-ins is required to play the audio files provided with the Lecture Notes on the CD. Click on the appropriate plug-in on the list below, save it onto your local hard drive, and run it! The instalation program should plug it into your browser.

●

Shockwave Player Plug-In, Win95/NT

●

Shockwave Player Plug-In, Windows 3.11

●

Shockwave Player Plug-In, Mac 68k

●

Shockwave Player Plug-In, Mac PowerPC

file:///E|/INSTALL/PLUGINS.HTM [19/1/2003 1:37:04]

Footnotes

...numbers. Yes, this is a true story. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK6/FOOTNODE.HTM [19/1/2003 1:37:06]

Implementation Challenges

Next: Data Structures and Sorting Up: Introduction to Algorithms Previous: Exercises

Implementation Challenges . Which of them gives better-quality solutions 1. (*) Implement the two TSP heuristics of Section in practice? Can you devise a heuristic that works better than both of them? 2. (*) Describe exactly how to test whether a given set of tickets proves minimum coverage in the Lotto problem of Section

. Write a program to find good ticket sets.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE21.HTM [19/1/2003 1:37:07]

Implementation Challenges

Next: Breaking Problems Down Up: Data Structures and Sorting Previous: Exercises

Implementation Challenges 1. Implement versions of several different dictionary data structures, such as linked lists, binary trees, balanced binary search trees, and hash tables. Conduct experiments to assess the relative performance of these data structures in a simple application that reads a large text file and reports exactly one instance of each word that appears within it. This application can be efficiently implemented by maintaining a dictionary of all distinct words that have appeared thus far in the text and inserting/reporting each word that is not found. Write a brief report with your conclusions. 2. Implement versions of several different sorting algorithms, such as selection sort, insertion sort, heapsort, mergesort, and quicksort. Conduct experiments to assess the relative performance of these algorithms in a simple application that reads a large text file and reports exactly one instance of each word that appears within it. This application can be efficiently implemented by sorting all the words that occur in the text and then passing through the sorted sequence to identify one instance of each distinct word. Write a brief report with your conclusions.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK/NODE41.HTM [19/1/2003 1:37:08]

Implementation Challenges

Next: Graph Algorithms Up: Breaking Problems Down Previous: Exercises

Implementation Challenges 1. (*) Many types of products sold appeal more to members of one ethnic group than another. Perhaps Greeks eat more pasta per capita than Russians do, while Japanese find baseball more appealing than do Italians. A market researcher might be interested in having a program scan the names on a mailing list to select the ones most likely to be, say, Greek to target for a given mailing. Develop a program that makes reasonable mappings between pairs of first/ last names and ethnicities. One approach would be to compute the edit distance between query names and a family of names of known ethnicity. Feel free to experiment with other approaches. 2. (*) In the game of Battleship, the first player hides a collection of, say, three ships on a grid. The second player guesses a series of grid positions and is informed whether they hit or miss a battleship. The second player continues to query until each of the battleship positions has been probed. While the second player must succeed after making 100 different probes, we seek a strategy to use as few probes as possible to achieve the goal. Develop a program that tries to efficiently sink all the battleships. One reasonable algorithmic approach would be based on divide-and-conquer or binary search. 3. (*) A Caesar shift (see Section ) is a very simple class of ciphers for secret messages. Unfortunately, they can be broken using statistical properties of English. Develop a program capable of decrypting Caesar shifts of sufficiently long texts.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE58.HTM [19/1/2003 1:37:09]

Implementation Challenges

Next: Combinatorial Search and Heuristic Up: Graph Algorithms Previous: Exercises

Implementation Challenges 1. Airline flight schedules define a natural graph, where the vertices are the airports and there is an edge between any two airports with direct flights between them whose weight is proportional to the distance between them. An extensive airplane data set due to Roberto Tammasia is available from the Algorithm Repository WWW/CD-ROM. Write a program that explicitly constructs the airport graph from this data set. 2. This problem is a follow-up to the exercise above. Changing planes repeatedly on connecting flights can be a hassle. Develop and implement an algorithm that finds the fewest flights needed to get from airport A to B, regardless of waiting time. 3. (*) This problem is a follow-up to the exercise above. Develop and implement an algorithm that finds the flight plan from airport A to B that minimizes the total distance traveled. 4. (*) This problem is a follow-up to the exercise above. Suppose that we must arrive at airport B at time T for an important scheduled meeting. Develop and implement an algorithm that finds the latest time one can leave airport A in time to make the meeting. 5. (*) This problem is a follow-up to the exercise above. In order to take advantage of a frequent flyer program, we might want to fly only on a particular airline. How can we modify the above solutions so as to accommodate such queries? 6. (**) This problem is a follow-up to the exercise above. A really devout frequent flyer might want to find the longest flight plan between airports A and B, so as to maximize the number of miles they get credit for. Develop and implement an algorithm to find the longest such route.

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK2/NODE82.HTM [19/1/2003 1:37:10]

Implementation Challenges

Next: Intractable Problems and Approximations Up: Combinatorial Search and Heuristic Previous: Exercises

Implementation Challenges 1. (*) Anagrams are rearrangements of the letters of a word or phrase into a different word or phrase. Sometimes the results are quite striking. For example, ``MANY VOTED BUSH RETIRED'' is an anagram of ``TUESDAY NOVEMBER THIRD,'' which correctly predicted the outcome of the 1992 U.S. presidential election. Design and implement an algorithm for finding anagrams using combinatorial search and a dictionary. 2. (*) For any of the exercises above, design and implement a simulated annealing heuristic to get reasonable solutions. How well does your program perform in practice? 3. (**) Design and implement a parallel sorting algorithm that distributes data across several processors. An appropriate variation of mergesort is a likely candidate. Measure the speedup of this algorithm as the number of processors increases. Later, compare the execution time to that of a purely sequential mergesort implementation. What are your experiences?

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE103.HTM [19/1/2003 1:37:11]

Implementation Challenges

Next: How to Design Algorithms Up: Intractable Problems and Approximations Previous: Exercises

Implementation Challenges 1. Implement a translator that translates satisfiability instances into equivallent 3-SAT instances. 2. (*) Design and implement a backtracking algorithm to test whether a set of formulae are satisfiable. What criteria can you use to prune this search? 3. (*) Implement the vertex cover to satisfiability reduction, and run the resulting clauses through a satisfiability testing code. Does this seem like a practical way to compute things?

Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE123.HTM [19/1/2003 1:37:12]

Caveats

Next: Data Structures Up: A Catalog of Algorithmic Previous: A Catalog of Algorithmic

Caveats This is a catalog of algorithmic problems. It is not a cookbook. It cannot be because there are too many possible recipes and too many possible variations on what those want to eat. My goal is to aim you in the right direction so that you can solve your own problems. I point out the issues you are likely to encounter along the way, problems that you are going to have to work out for yourself. In particular: ●

●

●

●

For each problem, I suggest algorithms and directions to attack it. These recommendations are based on my experiences and study and aimed towards what I see as typical applications. I felt it was more important to make concrete recommendations for the masses rather than to try to cover all possible situations. If you don't want to follow my advice, don't follow my advice. But before you ignore me, make sure you understand the reasoning behind my recommendations and can articulate a reason why your application violates my assumptions. The implementations I recommend are not necessarily complete solutions to your problem. I point to an implementation whenever I feel it might be more useful to someone than just a textbook description of the algorithm. Some programs are useful only as models for you to write your own codes. Others are embedded in large systems and so might be too painful to extract and run on their own. Assume that all of them contain bugs. Many are quite serious, so beware. Please respect the licensing conditions for any implementations you use commercially. Many of for a further these codes are not in the public domain and have restrictions. See Section discussion of this issue. I would be interested in hearing about your experiences with my recommendations, both positive and negative. I would be especially interested in learning about any other implementations that you know about. Feel free to drop me a line at [email protected].

Next: Data Structures Up: A Catalog of Algorithmic Previous: A Catalog of Algorithmic Algorithms Mon Jun 2 23:33:50 EDT 1997

file:///E|/BOOK/BOOK3/NODE127.HTM [19/1/2003 1:37:12]

1.1.1 Dictionaries

1.1.1 Dictionaries

INPUT

OUTPUT

Input Description: A set of n records, each identified by one or more key fields. Problem: Build and maintain a data structure to efficiently locate, insert, or delete the record associated with any query key q .

Implementations ● ● ● ●

LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 10) Handbook of Algorithms and Data Structures (Pascal) (rating 8) Xtango and Polka Algorithm Animation Systems (C++) (rating 5) DIMACS Implementation Challenges (FORTRAN) (rating 1)

Related Problems file:///E|/WEBSITE/FILES/DICTRIES.HTM (1 of 2) [19/1/2003 1:37:13]

1.1.1 Dictionaries ● ● ●

Priority Queues Searching Sorting

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/DICTRIES.HTM (2 of 2) [19/1/2003 1:37:13]

1.1.4 Graph Data Structures

1.1.4 Graph Data Structures

INPUT

OUTPUT

Input Description: A graph G . Problem: Give an efficient, flexible data structure to represent G .

Implementations ● ● ● ● ● ●

LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 10) GraphEd -- Graph Editor and Layout Program (C) (rating 8) Combinatorica (Mathematica) (rating 6) The Stanford GraphBase (C) (rating 6) LINK -- Programming and Visualization Environment for Hypergraphs (C++) (rating 6) Moret and Shapiro's Algorithms P to NP (Pascal) (rating 4)

Related Problems file:///E|/WEBSITE/FILES/GRAPURES.HTM (1 of 2) [19/1/2003 1:37:14]

1.1.4 Graph Data Structures

● ●

Graph Partition Set Data Structures

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/GRAPURES.HTM (2 of 2) [19/1/2003 1:37:14]

1.1.5 Set Data Structures

1.1.5 Set Data Structures

INPUT

OUTPUT

Input Description: A universe of objects U = \{ u_1,...,u_n\} , and a collection of subsets S_1,...,S_m , S_i \subset U . Problem: Represent each subset so as to efficiently (1) test whether u_i \in S_j , (2) find the union or intersection of S_i and S_j , (3) insert or delete members of S_i .

Implementations ● ● ●

LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 5) LINK -- Programming and Visualization Environment for Hypergraphs (C++) (rating 4) Moret and Shapiro's Algorithms P to NP (Pascal) (rating 4)

Related Problems file:///E|/WEBSITE/FILES/SET_URES.HTM (1 of 2) [19/1/2003 1:37:14]

1.1.5 Set Data Structures ● ● ● ● ●

Generating Partitions Generating Subsets Graph Data Structures Minimum Spanning Tree Set Cover

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/SET_URES.HTM (2 of 2) [19/1/2003 1:37:14]

1.2.1 Solving Linear Equations

1.2.1 Solving Linear Equations

INPUT

OUTPUT

Input Description: An m x n matrix A , and an m x 1 vector b , representing m linear equations with n variables. Problem: What is the vector x such that A \cdot x = b ?

Implementations ● ● ●

LAPACK and LINPACK -- Linear Algebra PACKages (FORTRAN) (rating 10) Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 7) LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 3)

Related Problems ●

Bandwidth Reduction file:///E|/WEBSITE/FILES/LINEIONS.HTM (1 of 2) [19/1/2003 1:37:15]

1.2.1 Solving Linear Equations ● ●

Determinants and Permanents Matrix Multiplication

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/LINEIONS.HTM (2 of 2) [19/1/2003 1:37:15]

1.2.2 Bandwidth Reduction

1.2.2 Bandwidth Reduction

INPUT

OUTPUT

Input Description: A graph G=(V,E) , representing an n x n matrix M of zero and non-zero elements. Problem: Which permutation p of the vertices of V minimizes \max_{(i,j) \in E} |p(i) - p(j)| , or equivalently the length of the longest edge when the vertices are ordered on a line.

Implementations ● ●

Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 9) Stony Brook Project Implementations (C++) (rating 6)

Related Problems ● ●

Feedback Edge/Vertex Set Solving Linear Equations file:///E|/WEBSITE/FILES/BANDIDTH.HTM (1 of 2) [19/1/2003 1:37:15]

1.2.2 Bandwidth Reduction ●

Topological Sorting

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/BANDIDTH.HTM (2 of 2) [19/1/2003 1:37:15]

1.2.4 Determinants and Permanents

1.2.4 Determinants and Permanents

INPUT

OUTPUT

Input Description: An n x n matrix M . Problem: What is the determinant |M| or the permanent Perm(M) for matrix m ?

Implementations ● ● ●

Combinatorica (Mathematica) (rating 4) Nijenhuis and Wilf: Combinatorial Algorithms (FORTRAN) (rating 4) LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 2)

Related Problems ●

Robust Geometric Primitives

file:///E|/WEBSITE/FILES/DETEANTS.HTM (1 of 2) [19/1/2003 1:37:16]

1.2.4 Determinants and Permanents ● ●

Solving Linear Equations Matching

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/DETEANTS.HTM (2 of 2) [19/1/2003 1:37:16]

1.2.8 Factoring and Primality Testing

1.2.8 Factoring and Primality Testing

INPUT

Input Description: An integer n . Problem: Is n a prime number, and if not what are the factors of n ?

Implementations ●

PARI - Package for Number Theory (C) (rating 9)

Related Problems ● ●

Cryptography Arbitrary Precision Arithmetic

file:///E|/WEBSITE/FILES/FACTGERS.HTM (1 of 2) [19/1/2003 1:37:16]

OUTPUT

1.2.8 Factoring and Primality Testing

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/FACTGERS.HTM (2 of 2) [19/1/2003 1:37:16]

1.2.9 Arbitrary Precision Arithmetic

1.2.9 Arbitrary Precision Arithmetic

INPUT

OUTPUT

Input Description: Two very large integers, x and y . Problem: What is x+y , x-y , x x y and x / y ?

Implementations ● ● ● ● ●

PARI - Package for Number Theory (C) (rating 9) Handbook of Algorithms and Data Structures (Pascal) (rating 6) LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 5) Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 5) Algorithms in C++ -- Sedgewick (C++) (rating 3)

Related Problems ● ●

Calendrical Calculations Cryptography file:///E|/WEBSITE/FILES/HIGHETIC.HTM (1 of 2) [19/1/2003 1:37:17]

1.2.9 Arbitrary Precision Arithmetic ● ●

Factoring and Primality Testing Discrete Fourier Transform

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/HIGHETIC.HTM (2 of 2) [19/1/2003 1:37:17]

1.2.10 Knapsack Problem

1.2.10 Knapsack Problem

INPUT

OUTPUT

Input Description: A set of items S=\{1,...,n\} , where item i has size s_i and value v_i . A knapsack capacity C . Problem: Find the subset S' \subset S which maximizes the value of \sum_{i \in S'} v_i given that \sum_{i \in S'} s_i \leq C , ie. fits in a knapsack of size C .

Implementations ● ●

Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 6) Discrete Optimization Methods (Pascal) (rating 4)

Related Problems ●

Bin Packing file:///E|/WEBSITE/FILES/00000017.HTM (1 of 2) [19/1/2003 1:37:18]

1.2.10 Knapsack Problem ●

Linear Programming

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/00000017.HTM (2 of 2) [19/1/2003 1:37:18]

1.3.2 Searching

1.3.2 Searching

INPUT

OUTPUT

Input Description: A set S of n keys, a query key q . Problem: Where is q in S ?

Implementations ● ● ●

LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 7) Handbook of Algorithms and Data Structures (Pascal) (rating 7) Moret and Shapiro's Algorithms P to NP (Pascal) (rating 3)

Related Problems ●

Dictionaries

file:///E|/WEBSITE/FILES/SEARHING.HTM (1 of 2) [19/1/2003 1:37:18]

1.3.2 Searching ●

Sorting

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/SEARHING.HTM (2 of 2) [19/1/2003 1:37:18]

1.3.4 Generating Permutations

1.3.4 Generating Permutations

INPUT

OUTPUT

Input Description: An integer n . Problem: Generate (1) all, or (2) a random, or (3) the next permutation of length n .

Implementations ● ● ● ●

Frank Ruskey's Combinatorial Generation Resources (Pascal) (rating 8) Nijenhuis and Wilf: Combinatorial Algorithms (FORTRAN) (rating 8) Combinatorica (Mathematica) (rating 7) The Stanford GraphBase (C) (rating 5)

Related Problems file:///E|/WEBSITE/FILES/GENTIONS.HTM (1 of 2) [19/1/2003 1:37:19]

1.3.4 Generating Permutations ● ● ● ● ●

Calendrical Calculations Generating Graphs Generating Partitions Generating Subsets Random Number Generation

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/GENTIONS.HTM (2 of 2) [19/1/2003 1:37:19]

1.3.5 Generating Subsets

1.3.5 Generating Subsets

INPUT

OUTPUT

Input Description: An integer n . Problem: Generate (1) all, or (2) a random, or (3) the next subset of the integers 1 to n .

Implementations ● ● ● ●

Nijenhuis and Wilf: Combinatorial Algorithms (FORTRAN) (rating 8) Combinatorica (Mathematica) (rating 7) Frank Ruskey's Combinatorial Generation Resources (Pascal) (rating 7) Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 3)

Related Problems file:///E|/WEBSITE/FILES/GSUBSETS.HTM (1 of 2) [19/1/2003 1:37:19]

1.3.5 Generating Subsets ● ● ● ●

Generating Partitions Generating Permutations Random Number Generation Set Data Structures

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/GSUBSETS.HTM (2 of 2) [19/1/2003 1:37:19]

1.3.6 Generating Partitions

1.3.6 Generating Partitions

INPUT

OUTPUT

Input Description: An integer n . Problem: Generate (1) all, or (2) a random, or (3) the next integer or set partitions of length n .

Implementations ● ● ● ● ●

Nijenhuis and Wilf: Combinatorial Algorithms (FORTRAN) (rating 8) Combinatorica (Mathematica) (rating 7) Frank Ruskey's Combinatorial Generation Resources (Pascal) (rating 7) Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 6) The Stanford GraphBase (C) (rating 5)

Related Problems file:///E|/WEBSITE/FILES/GEITIONS.HTM (1 of 2) [19/1/2003 1:37:20]

1.3.6 Generating Partitions ● ● ● ●

Generating Permutations Generating Subsets Random Number Generation Set Data Structures

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/GEITIONS.HTM (2 of 2) [19/1/2003 1:37:20]

1.3.8 Calendrical Calculations

1.3.8 Calendrical Calculations

INPUT

OUTPUT

Input Description: A particular calendar date d , specified by month, day and year. Problem: Which day of the week did d fall on in the given calendar system?

Implementations ●

Calendrical Calculations (C++) (rating 10)

Related Problems ● ●

Generating Permutations Arbitrary Precision Arithmetic

file:///E|/WEBSITE/FILES/CALEIONS.HTM (1 of 2) [19/1/2003 1:37:21]

1.3.8 Calendrical Calculations

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/CALEIONS.HTM (2 of 2) [19/1/2003 1:37:21]

1.3.9 Job Scheduling

1.3.9 Job Scheduling

INPUT

OUTPUT

Input Description: A directed acyclic graph G=(V,E) , where the vertices represent jobs and the the edge (u,v) that task u must be completed before task v . Problem: What schedule of tasks to completes the job using the minimum amount of time or processors?

Implementations ● ●

Discrete Optimization Methods (Pascal) (rating 4) Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 4)

Related Problems ● ●

Bin Packing Edge Coloring file:///E|/WEBSITE/FILES/SCHELING.HTM (1 of 2) [19/1/2003 1:37:21]

1.3.9 Job Scheduling ● ● ● ●

Feedback Edge/Vertex Set Matching Topological Sorting Vertex Coloring

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/SCHELING.HTM (2 of 2) [19/1/2003 1:37:21]

1.4.2 Topological Sorting

1.4.2 Topological Sorting

INPUT

OUTPUT

Input Description: A directed, acyclic graph G=(V,E) (also known as a partial order or poset). Problem: Find a linear ordering of the vertices of V such that for each edge (i,j) \in E , vertex i is to the left of vertex j .

Implementations ● ● ● ● ● ●

LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 7) Combinatorica (Mathematica) (rating 3) The Stanford GraphBase (C) (rating 3) Moret and Shapiro's Algorithms P to NP (Pascal) (rating 3) Algorithms in C++ -- Sedgewick (C++) (rating 3) Xtango and Polka Algorithm Animation Systems (C++) (rating 2)

file:///E|/WEBSITE/FILES/TOPOTING.HTM (1 of 2) [19/1/2003 1:37:22]

1.4.2 Topological Sorting

Related Problems ● ● ● ●

Bandwidth Reduction Feedback Edge/Vertex Set Job Scheduling Sorting

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/TOPOTING.HTM (2 of 2) [19/1/2003 1:37:22]

1.4.8 Edge and Vertex Connectivity

1.4.8 Edge and Vertex Connectivity

INPUT

OUTPUT

Input Description: A graph G . Optionally, a pair of vertices and t . Problem: What is the smallest subset of vertices (edges) whose deletion will disconnect G ? Alternately, what is the smallest subset of vertices (edges) which will separate from t ?

Implementations ● ● ●

Combinatorica (Mathematica) (rating 4) The Stanford GraphBase (C) (rating 4) Moret and Shapiro's Algorithms P to NP (Pascal) (rating 4)

Related Problems ●

Connected Components file:///E|/WEBSITE/FILES/EDGEVITY.HTM (1 of 2) [19/1/2003 1:37:22]

1.4.8 Edge and Vertex Connectivity ● ●

Graph Partition Network Flow

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/EDGEVITY.HTM (2 of 2) [19/1/2003 1:37:22]

1.4.10 Drawing Graphs Nicely

1.4.10 Drawing Graphs Nicely

INPUT

Input Description: A graph G . Problem: Give a drawing of graph G which accurately reflects its structure.

Implementations ● ● ●

GraphEd -- Graph Editor and Layout Program (C) (rating 9) GraphViz -- graph layout programs (C) (rating 7) Combinatorica (Mathematica) (rating 6)

Related Problems ●

Drawing Trees

file:///E|/WEBSITE/FILES/DRARAPHS.HTM (1 of 2) [19/1/2003 1:37:23]

OUTPUT

1.4.10 Drawing Graphs Nicely ●

Planarity Detection and Embedding

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/DRARAPHS.HTM (2 of 2) [19/1/2003 1:37:23]

1.4.11 Drawing Trees

1.4.11 Drawing Trees

INPUT

Input Description: A tree (ie. graph without any cycles) T . Problem: A nice drawing of the tree T .

Implementations ● ● ●

GraphEd -- Graph Editor and Layout Program (C) (rating 9) GraphViz -- graph layout programs (C) (rating 7) Combinatorica (Mathematica) (rating 6)

Related Problems ●

Drawing Graphs Nicely

file:///E|/WEBSITE/FILES/DRATREES.HTM (1 of 2) [19/1/2003 1:37:23]

OUTPUT

1.4.11 Drawing Trees ●

Planarity Detection and Embedding

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/DRATREES.HTM (2 of 2) [19/1/2003 1:37:23]

1.4.12 Planarity Detection and Embedding

1.4.12 Planarity Detection and Embedding

INPUT

OUTPUT

Input Description: A graph G . Problem: Can G be drawn in the plane such that no two edges cross? If so, produce such a drawing.

Implementations ● ● ●

GraphEd -- Graph Editor and Layout Program (C) (rating 8) LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 7) Combinatorica (Mathematica) (rating 2)

Related Problems ●

Drawing Graphs Nicely

file:///E|/WEBSITE/FILES/PLANWING.HTM (1 of 2) [19/1/2003 1:37:24]

1.4.12 Planarity Detection and Embedding ● ●

Drawing Trees Graph Partition

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/PLANWING.HTM (2 of 2) [19/1/2003 1:37:24]

1.5.2 Independent Set

1.5.2 Independent Set

INPUT

OUTPUT

Input Description: A graph G=(V,E) . Problem: What is the largest subset of vertices of V such that no pair of vertices defines an edge of E ?

Implementations ● ●

DIMACS Implementation Challenges (FORTRAN) (rating 7) Neural-Networks for Cliques and Coloring (C) (rating 5)

Related Problems ● ●

Clique Set Packing

file:///E|/WEBSITE/FILES/INDE_SET.HTM (1 of 2) [19/1/2003 1:37:25]

1.5.2 Independent Set ● ●

Vertex Coloring Vertex Cover

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/INDE_SET.HTM (2 of 2) [19/1/2003 1:37:25]

1.5.3 Vertex Cover

1.5.3 Vertex Cover

INPUT

OUTPUT

Input Description: A graph G=(V,E) . Problem: What is the smallest subset S \subset V such that each e \in E contains at least one vertex of S ?

Implementations ● ● ●

Neural-Networks for Cliques and Coloring (C) (rating 6) DIMACS Implementation Challenges (FORTRAN) (rating 5) Combinatorica (Mathematica) (rating 4)

Related Problems ●

Clique

file:///E|/WEBSITE/FILES/VERCOVER.HTM (1 of 2) [19/1/2003 1:37:25]

1.5.3 Vertex Cover ● ●

Independent Set Set Cover

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/VERCOVER.HTM (2 of 2) [19/1/2003 1:37:25]

1.5.4 Traveling Salesman Problem

1.5.4 Traveling Salesman Problem

INPUT

OUTPUT

Input Description: A weighted graph G . Problem: Find the cycle of minimum cost visiting all of the vertices of G exactly once.

Implementations ● ● ● ● ●

TSP solvers (C) (rating 8) Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 6) Discrete Optimization Methods (Pascal) (rating 5) Combinatorica (Mathematica) (rating 3) Xtango and Polka Algorithm Animation Systems (C++) (rating 3)

Related Problems file:///E|/WEBSITE/FILES/TRAESMAN.HTM (1 of 2) [19/1/2003 1:37:26]

1.5.4 Traveling Salesman Problem ● ● ● ●

Convex Hull Hamiltonian Cycle Minimum Spanning Tree Satisfiability

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/TRAESMAN.HTM (2 of 2) [19/1/2003 1:37:26]

1.5.6 Graph Partition

1.5.6 Graph Partition

INPUT

OUTPUT

Input Description: A (weighted) graph G=(V,E) . Integers j , k , and m . Problem: Partition the vertices into m subsets such that each subset has size at most j , while the cost of the edges spanning subsets is bounded by k .

Implementations ● ●

LINK -- Programming and Visualization Environment for Hypergraphs (C++) (rating 8) LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 4)

Related Problems ● ●

Edge and Vertex Connectivity Graph Data Structures file:///E|/WEBSITE/FILES/GRAPTION.HTM (1 of 2) [19/1/2003 1:37:26]

1.5.6 Graph Partition ● ●

Network Flow Planarity Detection and Embedding

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/GRAPTION.HTM (2 of 2) [19/1/2003 1:37:26]

1.5.7 Vertex Coloring

1.5.7 Vertex Coloring

INPUT

OUTPUT

Input Description: A graph G=(V,E) . Problem: Color the vertices of V with the minimum number of colors such that for each edge (i,j) \in E , vertices i and j have different colors.

Implementations ● ● ● ● ● ● ● ●

DIMACS Implementation Challenges (FORTRAN) (rating 7) Mike Trick's Graph Coloring Resources (C) (rating 7) Neural-Networks for Cliques and Coloring (C) (rating 6) Joe Culberson's Graph Coloring Resources (C) (rating 6) Discrete Optimization Methods (Pascal) (rating 4) Xtango and Polka Algorithm Animation Systems (C++) (rating 4) Combinatorica (Mathematica) (rating 3) Nijenhuis and Wilf: Combinatorial Algorithms (FORTRAN) (rating 3) file:///E|/WEBSITE/FILES/VERTRING.HTM (1 of 2) [19/1/2003 1:37:27]

1.5.7 Vertex Coloring

Related Problems ● ● ●

Edge Coloring Independent Set Job Scheduling

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES/VERTRING.HTM (2 of 2) [19/1/2003 1:37:27]

1.5.11 Feedback Edge/Vertex Set

1.5.11 Feedback Edge/Vertex Set

INPUT

OUTPUT

Input Description: A (directed) graph G=(V,E) . Problem: What is the smallest set of edges E' or vertices V' whose deletion leaves an acyclic graph?

Implementations ●

The Stanford GraphBase (C) (rating 4)

Related Problems ● ● ●

Bandwidth Reduction Job Scheduling Topological Sorting

file:///E|/WEBSITE/FILES2/FEED_SET.HTM (1 of 2) [19/1/2003 1:37:27]

1.5.11 Feedback Edge/Vertex Set

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/FEED_SET.HTM (2 of 2) [19/1/2003 1:37:27]

1.6.1 Robust Geometric Primitives

1.6.1 Robust Geometric Primitives

INPUT

OUTPUT

Implementations ● ● ● ● ●

LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 8) Joseph O'Rourke's Computational Geometry (C) (rating 6) Geolab -- Computational Geometry System (C++) (rating 5) Moret and Shapiro's Algorithms P to NP (Pascal) (rating 4) Algorithms in C++ -- Sedgewick (C++) (rating 4)

Related Problems ● ● ●

Determinants and Permanents Intersection Detection Maintaining Line Arrangements

file:///E|/WEBSITE/FILES2/GEOMIVES.HTM (1 of 2) [19/1/2003 1:37:29]

1.6.1 Robust Geometric Primitives

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/GEOMIVES.HTM (2 of 2) [19/1/2003 1:37:29]

1.6.6 Range Search

1.6.6 Range Search

INPUT

OUTPUT

Input Description: A set S of n points in E^d , and a query polygon Q . Problem: Which points from S lie within Q ?

Implementations ● ● ● ●

LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 7) Ranger - Nearest Neighbor Search in Higher Dimensions (C) (rating 6) Algorithms in C++ -- Sedgewick (C++) (rating 4) Handbook of Algorithms and Data Structures (Pascal) (rating 2)

Related Problems file:///E|/WEBSITE/FILES2/RANGARCH.HTM (1 of 2) [19/1/2003 1:37:30]

1.6.6 Range Search ● ● ●

Kd-Trees Nearest Neighbor Search Point Location

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/RANGARCH.HTM (2 of 2) [19/1/2003 1:37:30]

1.6.8 Intersection Detection

1.6.8 Intersection Detection

INPUT

OUTPUT

Input Description: A set S of lines and line segments l_1,...,l_n , or a pair of polygons or polyhedra P_1 and P_2 . Problem: Which pairs of line segments intersect each other? What is the intersection of P_1 and P_2 ?

Implementations ● ● ● ● ●

LEDA - A Library of Efficient Data Types and Algorithms (C++) (rating 7) Joseph O'Rourke's Computational Geometry (C) (rating 5) Qhull - higher dimensional convex hull program (C) (rating 5) Moret and Shapiro's Algorithms P to NP (Pascal) (rating 3) Xtango and Polka Algorithm Animation Systems (C++) (rating 1)

Related Problems file:///E|/WEBSITE/FILES2/INTETION.HTM (1 of 2) [19/1/2003 1:37:30]

1.6.8 Intersection Detection

● ● ●

Robust Geometric Primitives Maintaining Line Arrangements Motion Planning

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/INTETION.HTM (2 of 2) [19/1/2003 1:37:30]

1.6.16 Minkowski Sum

1.6.16 Minkowski Sum

INPUT

OUTPUT

Input Description: Point sets or polygons A and B , with n and m vertices, respectively. Problem: What is the convolution of A and B , ie. the Minkowski sum A+B = \{x+y| x\in A, y \in B\} ?

Implementations ●

David Eppstein's Knuth-Morris-Pratt Algorithm and Minkowski sum code (C++) (rating 4)

Related Problems ● ● ●

Motion Planning Simplifying Polygons Medial-Axis Transformation

file:///E|/WEBSITE/FILES2/MINK_SUM.HTM (1 of 2) [19/1/2003 1:37:31]

1.6.16 Minkowski Sum

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/MINK_SUM.HTM (2 of 2) [19/1/2003 1:37:31]

1.7.1 Set Cover

1.7.1 Set Cover

INPUT

OUTPUT

Input Description: A set of subsets S_1, ..., S_m of the universal set U = \{1,...,n\} . Problem: What is the smallest subset of subsets T \subset S such that \cup_{t_i \in T} t_i = U ?

Implementations ●

Discrete Optimization Methods (Pascal) (rating 5)

Related Problems ● ● ●

Matching Polygon Partitioning Set Data Structures

file:///E|/WEBSITE/FILES2/SET_OVER.HTM (1 of 2) [19/1/2003 1:37:32]

1.7.1 Set Cover ● ●

Set Packing Vertex Cover

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/SET_OVER.HTM (2 of 2) [19/1/2003 1:37:32]

1.7.2 Set Packing

1.7.2 Set Packing

INPUT

OUTPUT

Input Description: A set of subsets S = S_1, ..., S_m of the universal set U = \{1,...,n\} . Problem: What is the largest number of mutually disjoint subsets from S ?

Implementations ●

Discrete Optimization Methods (Pascal) (rating 5)

Related Problems ● ● ●

Bin Packing Independent Set Set Cover

file:///E|/WEBSITE/FILES2/SET_KING.HTM (1 of 2) [19/1/2003 1:37:32]

1.7.2 Set Packing

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/SET_KING.HTM (2 of 2) [19/1/2003 1:37:32]

1.7.3 String Matching

1.7.3 String Matching

INPUT

OUTPUT

Input Description: A text string t of length n . A patterns string p of length m . Problem: Find the first (or all) instances of the pattern in the text.

Implementations ● ● ● ● ●

Fire-Engine and Spare-Parts String and Language Algorithms (C++) (rating 7) David Eppstein's Knuth-Morris-Pratt Algorithm and Minkowski sum code (C++) (rating 4) Handbook of Algorithms and Data Structures (Pascal) (rating 4) Algorithms in C++ -- Sedgewick (C++) (rating 4) Xtango and Polka Algorithm Animation Systems (C++) (rating 3)

Related Problems file:///E|/WEBSITE/FILES2/STRIHING.HTM (1 of 2) [19/1/2003 1:37:33]

1.7.3 String Matching ● ● ● ●

Approximate String Matching Finite State Machine Minimization Graph Isomorphism Suffix Trees and Arrays

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/STRIHING.HTM (2 of 2) [19/1/2003 1:37:33]

1.7.4 Approximate String Matching

1.7.4 Approximate String Matching

INPUT

OUTPUT

Input Description: A text string t and a pattern string p . An edit cost bound k . Problem: Does there exist an alignment between t and p with edit cost at most k , ie. can we transform part of t to p using at most k additions, deletions, and substitutions.

Implementations ● ● ●

agrep - Approximate General Regular Expression Pattern Matcher (C) (rating 10) HT/DIG -- image compression codes (C) (rating 7) Handbook of Algorithms and Data Structures (Pascal) (rating 2)

Related Problems ●

Longest Common Substring file:///E|/WEBSITE/FILES2/APPRHING.HTM (1 of 2) [19/1/2003 1:37:33]

1.7.4 Approximate String Matching ●

String Matching

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/APPRHING.HTM (2 of 2) [19/1/2003 1:37:33]

1.7.5 Text Compression

1.7.5 Text Compression

INPUT

OUTPUT

Input Description: A text string S . Problem: A shortest text string S' such that S can be reconstructed from S' .

Implementations ●

Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 5)

Related Problems ● ● ●

Cryptography Discrete Fourier Transform Shortest Common Superstring

file:///E|/WEBSITE/FILES2/TEXSSION.HTM (1 of 2) [19/1/2003 1:37:34]

1.7.5 Text Compression

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/TEXSSION.HTM (2 of 2) [19/1/2003 1:37:34]

1.7.6 Cryptography

1.7.6 Cryptography

INPUT

OUTPUT

Input Description: A plaintext message T or encrypted text E , and a key k . Problem: Encode T using k giving E , or decode E using k back to T .

Implementations ● ● ●

() (rating 10) The Stanford GraphBase (C) (rating 3) Netlib / TOMS -- Collected Algorithms of the ACM (FORTRAN) (rating 3)

Related Problems ● ●

Factoring and Primality Testing Arbitrary Precision Arithmetic file:///E|/WEBSITE/FILES2/CRYPAPHY.HTM (1 of 2) [19/1/2003 1:37:34]

1.7.6 Cryptography ●

Text Compression

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/CRYPAPHY.HTM (2 of 2) [19/1/2003 1:37:34]

1.7.7 Finite State Machine Minimization

1.7.7 Finite State Machine Minimization

INPUT

OUTPUT

Input Description: A deterministic finite automata M . Problem: The smallest deterministic finite automata M' such that M' behaves identically to M'

Implementations ● ● ● ●

Grail: finite automata and regular expressions (C++) (rating 9) Fire-Engine and Spare-Parts String and Language Algorithms (C++) (rating 8) Handbook of Algorithms and Data Structures (Pascal) (rating 5) Xtango and Polka Algorithm Animation Systems (C++) (rating 1)

Related Problems file:///E|/WEBSITE/FILES2/FINITION.HTM (1 of 2) [19/1/2003 1:37:35]

1.7.7 Finite State Machine Minimization ● ●

Satisfiability String Matching

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/FINITION.HTM (2 of 2) [19/1/2003 1:37:35]

1.7.8 Longest Common Substring

1.7.8 Longest Common Substring

INPUT

OUTPUT

Input Description: A set S of strings S_1,...,S_n . Problem: What is the longest string c such that for each S_i , 1 \leq i \leq n , the characters of c appear as a scattered subsequence of S_i ?

Implementations ● ●

CAP -- Contig Assembly Program (C) (rating 8) Combinatorica (Mathematica) (rating 2)

Related Problems ● ●

Approximate String Matching Shortest Common Superstring file:///E|/WEBSITE/FILES2/LONGRING.HTM (1 of 2) [19/1/2003 1:37:35]

1.7.8 Longest Common Substring ●

Suffix Trees and Arrays

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/LONGRING.HTM (2 of 2) [19/1/2003 1:37:35]

1.7.9 Shortest Common Superstring

1.7.9 Shortest Common Superstring

INPUT

OUTPUT

Input Description: A set of strings s_1, ..., s_m . Problem: Find the shortest string S which contains each s_i as a substring of S .

Implementations ●

CAP -- Contig Assembly Program (C) (rating 8)

Related Problems ● ● ●

Longest Common Substring Suffix Trees and Arrays Text Compression

file:///E|/WEBSITE/FILES2/SHORRING.HTM (1 of 2) [19/1/2003 1:37:36]

1.7.9 Shortest Common Superstring

Go to the corresponding chapter in the book About the Book Send us Mail Go to Main Page This page last modified on Tue Jun 03, 1997 .

file:///E|/WEBSITE/FILES2/SHORRING.HTM (2 of 2) [19/1/2003 1:37:36]

Handbook of Algorithms and Data Structures

Handbook of Algorithms and Data Structures The Handbook of Data Structures and Algorithms , by Gonnet and Baeza-Yates, provides a comprehensive reference on fundamental data structures for searching and priority queues, and algorithms for sorting and text searching. The book covers these relatively small number of topics comprehensively, presenting most of the major and minor variations which appear in the literature. Perusing the book makes one aware of the tremendous range of data structures which have been developed, and the intense level of analysis many of them have been subjected to. For each data structure or algorithm, a brief description is provided along with its asymptotic complexity and an extensive set of references. More distinctively, an implementation in C and/or Pascal is usually provided, along with experimental results comparing the performance of these implementations on a variety of distributions. The authors have created a very slick WWW site for the book, http://www.dcc.uchile.cl/~rbaeza/handbook/hbook.html , containing all the source code as well as the index and table of contents. This entire site has been replicated in the local distribution site , although this copy still contains ftp links to Chile for the algorithm implementations. However, all of these files are also available locally . These files are all copyrighted © Addison-Wesley Publishing Co. Inc. Since many of the elementary data structures and sorting algorithms can be implemented concisely, most of the programs are very short. They are perhaps most useful as models, or as part of an experimental study to compare the performance on different data structures for a particular application.

● ● ●

Link to Handbook of Algorithms and Data Structures Local copies of all the algorithm implementations Local copy of original site, but with FTP calls to Chile for certain algorithm implementations

Problem Links ● ● ● ● ● ● ● ●

Dictionaries (8) Priority Queues (7) Searching (7) Sorting (7) Arbitrary Precision Arithmetic (6) Median and Selection (6) Finite State Machine Minimization (5) String Matching (4) file:///E|/WEBSITE/IMPLEMEN/HANDBOOK/IMPLEMEN.HTM (1 of 2) [19/1/2003 1:37:37]

Handbook of Algorithms and Data Structures ● ● ● ●

Kd-Trees (3) Approximate String Matching (2) Range Search (2) Suffix Trees and Arrays (2)

About the Book Send us Mail Go to Main Page This page last modified on May 21, 1997.

file:///E|/WEBSITE/IMPLEMEN/HANDBOOK/IMPLEMEN.HTM (2 of 2) [19/1/2003 1:37:37]

Moret and Shapiro's Algorithms P to NP

Moret and Shapiro's Algorithms P to NP This algorithms text distinguishes itself by including Pascal implementations of many algorithms, with careful experimental comparisons of different algorithms for such problems as sorting and minimum spanning tree, and heuristics for the traveling salesman problem. It provides a useful model for how to properly do empirical algorithm analysis. The programs themselves are probably best used as models. Interesting implementations include the eight-queens problem, fundamental graph and geometric algorithms. The programs in the book have been made available by anonymous ftp from cs.unm.edu in directory /pub/moret_shapiro.

● ●

Link to Moret's Home Page - files under publications Download Files (local site)

Problem Links ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

Sorting (7) Minimum Spanning Tree (5) Connected Components (4) Edge and Vertex Connectivity (4) Discrete Fourier Transform (4) Robust Geometric Primitives (4) Graph Data Structures (4) Matching (4) Network Flow (4) Set Data Structures (4) Convex Hull (3) Intersection Detection (3) Nearest Neighbor Search (3) Point Location (3) Priority Queues (3) Searching (3) Topological Sorting (3)

file:///E|/WEBSITE/IMPLEMEN/MORET/IMPLEMEN.HTM (1 of 2) [19/1/2003 1:37:38]

Moret and Shapiro's Algorithms P to NP

About the Book Send us Mail Go to Main Page This page last modified on Feb 23, 1996.

file:///E|/WEBSITE/IMPLEMEN/MORET/IMPLEMEN.HTM (2 of 2) [19/1/2003 1:37:38]

Index A

Up: Index - All

Index: A abstract data types adjacency list adjacency matrix algorithm, what is a all-pairs shortest path approximate string matching approximation algorithms articulation vertex asymptotic notation average-case analysis average-case complexity

Algorithms Mon Jun 2 14:49:38 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_A.HTM [19/1/2003 1:37:38]

Index B

Up: Index - All

Index: B back edge backsubstitution , backtracking backtracking - recursive backtracking contest , balanced search tree bandersnatch problem bandwidth - optimizations bandwidth problem bar codes - high density base of logarithm best-case complexity , big Oh notation , binary heap , binary search , binary search tree bin packing bipartite graphs birthday paradox bitonic tours bo-billy problem boundary condition , boundary conditions - string matching breadth-first search , breadth-first search - shortest paths breadth-first search - tree Brooks, Mel bubblesort - parallel bucketsort

file:///E|/LEC/LECTUR16/INDEX_B.HTM (1 of 2) [19/1/2003 1:37:39]

Index B

Algorithms Mon Jun 2 14:49:38 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_B.HTM (2 of 2) [19/1/2003 1:37:39]

Index C

Up: Index - All

Index: C chaining checkers chessboard covering chromatic number circuit board assembly clique clique - hardness closest pair heuristic closest pair problem clustering coloring graphs combinatorial search communications problems comparison-based sorting component design computer graphics connected components connected graph connectivity convex hull , Cook's theorem correctness crime cross edge cycle ,

Algorithms Mon Jun 2 14:49:38 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_C.HTM [19/1/2003 1:37:40]

Index D

Up: Index - All

Index: D DAG data structures decision problems , decision tree , degree delete - binary search tree delete - red-black tree depth-first search , depth-first search - edge organization deque diameter dictionary data structure , Dijkstra's algorithm , directed acyclic graph directed graph discrete event simulation disjunctive normal form divide-and-conquer , dominance relation duplicate elimination dynamic programming , dynamic programming - components dynamic programming - space dynamic programming - when applicable

Algorithms Mon Jun 2 14:49:38 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_D.HTM [19/1/2003 1:37:41]

Index E

Up: Index - All

Index: E edges edit distance element uniqueness exhaustive search algorithm exponential time

Algorithms Mon Jun 2 14:49:38 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_E.HTM [19/1/2003 1:37:41]

Index F

Up: Index - All

Index: F Federal sentencing guidelines Fibonacci numbers flat earth model Floyd-Warshall algorithm formal language theory forward edge Fourier transform fragment assembly frequency distribution friendship graph

Algorithms Mon Jun 2 14:49:38 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_F.HTM [19/1/2003 1:37:43]

Index G

Up: Index - All

Index: G gadget Garey and Johnson geometric algorithms , geometric series girth gossip graphs graph traversal greedy heuristic , guessing module

Algorithms Mon Jun 2 14:49:38 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_G.HTM [19/1/2003 1:37:43]

Index H

Up: Index - All

Index: H Hamiltonian cycle Hamiltonian cycle - hardness Hamiltonian path , harmonic numbers hash function hash table heap data structure heapify heapsort , Heep, Danny heuristic Huffman codes

Algorithms Mon Jun 2 14:49:39 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_H.HTM [19/1/2003 1:37:44]

Index I

Up: Index - All

Index: I incremental algorithms independent set induction , inorder traversal insert - binary search tree insert - red-black tree insertion sort , instance integer partition integer programming - hardness , inverse exponential function

Algorithms Mon Jun 2 14:49:39 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_I.HTM [19/1/2003 1:37:44]

Index K

Up: Index - All

Index: K Kruskal's algorithm ,

Algorithms Mon Jun 2 14:49:39 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_K.HTM [19/1/2003 1:37:45]

Index L

Up: Index - All

Index: L linked list local replacement logarithm logic problem longest increasing subsequence longest path lower bound lower bound - sorting

Algorithms Mon Jun 2 14:49:39 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_L.HTM [19/1/2003 1:37:45]

Index M

Up: Index - All

Index: M mainframe computers Maple master theorem , , Mathematica mathematical induction matrix multiplication , matrix multiplication - applications maze median mergesort minimum spanning tree , mode monotonic sequence morphing motion planning movie stars multigraph

Algorithms Mon Jun 2 14:49:39 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_M.HTM [19/1/2003 1:37:46]

Index N

Up: Index - All

Index: N nearest neighbor heuristic non-comparison based sorting non-deterministic Turning machine NP NP-completeness NP-completeness - dealing with it

Algorithms Mon Jun 2 14:49:39 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_N.HTM [19/1/2003 1:37:47]

Index O

Up: Index - All

Index: O open addressing optimization problem optimizing programs for speed order-statistic tree

Algorithms Mon Jun 2 14:49:39 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_O.HTM [19/1/2003 1:37:48]

Index P

Up: Index - All

Index: P parenthesization partial order partitioning step path path compression permutations - generation pivot pointers polygon triangulation polynomial time predecessor Prim's algorithm principle of optimality priority queue , problem , problem solving techniques , pseudocode P vs. NP ,

Algorithms Mon Jun 2 14:49:39 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_P.HTM [19/1/2003 1:37:48]

Index Q

Up: Index - All

Index: Q queens problem queue ,

,

quicksort , quicksort - analysis

Algorithms Mon Jun 2 14:49:39 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_Q.HTM [19/1/2003 1:37:49]

Index R

Up: Index - All

Index: R RAM model of computation random-access machine randomized algorithm , recurrence relations recursion tree , red-black tree red-black trees

,

reduction , reduction - direction of refrain refrigerator restriction robot arm Rolling Stones rotation rules for algorithm design

Algorithms Mon Jun 2 14:49:39 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_R.HTM [19/1/2003 1:37:49]

Index S

Up: Index - All

Index: S satisfiability satisfiability - Cook's theorem selection selection sort , sentence disambiguation sentinel series convergence set data structures , Shifflett shortest path , simple graph simulations social security number solving recurrences songs, complexity of sorting sorting - applications sorting - lower bound , space complexity spanning tree sparse graphs spouse, finding the optimal square of graph stack , , Strassen's algorithm string matching strongly connected components , subgraph isomorphism subset generation subset sum

file:///E|/LEC/LECTUR16/INDEX_S.HTM (1 of 2) [19/1/2003 1:37:50]

Index S

successor supercomputer , sweepline algorithms symmetry, exploiting

Algorithms Mon Jun 2 14:49:39 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_S.HTM (2 of 2) [19/1/2003 1:37:50]

Index T

Up: Index - All

Index: T telephone book telephone keypad telephone network , terrorist text compression topological sorting , transportation problems traveling salesman problem ,

,

,

,

traveling salesman problem - approximation , triangulation triangulation - minimum length Turing machine

Algorithms Mon Jun 2 14:49:39 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_T.HTM [19/1/2003 1:37:51]

Index U

Up: Index - All

Index: U undirected graph union-find data structure upper bound

Algorithms Mon Jun 2 14:49:39 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_U.HTM [19/1/2003 1:37:51]

Index V

Up: Index - All

Index: V vertex cover , vertex cover - approximation vertex cover - hardness vertex cover - tree vertices Viterbi algorithm

Algorithms Mon Jun 2 14:49:39 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_V.HTM [19/1/2003 1:37:52]

Index W

Up: Index - All

Index: W worst-case complexity

file:///E|/LEC/LECTUR16/INDEX_W.HTM [19/1/2003 1:37:52]

Index (complete)

Up: Index - All

Index: (complete) 3-satisfiability,

abstract data types adjacency list adjacency matrix algorithm, what is a all-pairs shortest path approximate string matching approximation algorithms articulation vertex asymptotic notation average-case analysis average-case complexity back edge backsubstitution, backtracking backtracking - recursive backtracking contest, balanced search tree bandersnatch problem bandwidth - optimizations bandwidth problem bar codes - high density base of logarithm best-case complexity, big Oh notation, binary heap, binary search, binary search tree file:///E|/LEC/LECTUR16/INDEX_AL.HTM (1 of 8) [19/1/2003 1:37:56]

Index (complete)

bin packing bipartite graphs birthday paradox bitonic tours bo-billy problem boundary condition, boundary conditions - string matching breadth-first search, breadth-first search - shortest paths breadth-first search - tree Brooks, Mel bubblesort - parallel bucketsort chaining checkers chessboard covering chromatic number circuit board assembly clique clique - hardness closest pair heuristic closest pair problem clustering coloring graphs combinatorial search communications problems comparison-based sorting component design computer graphics connected components connected graph connectivity convex hull, Cook's theorem correctness crime cross edge

file:///E|/LEC/LECTUR16/INDEX_AL.HTM (2 of 8) [19/1/2003 1:37:56]

Index (complete)

cycle, DAG data structures decision problems, decision tree, degree delete - binary search tree delete - red-black tree depth-first search, depth-first search - edge organization deque diameter dictionary data structure, Dijkstra's algorithm, directed acyclic graph directed graph discrete event simulation disjunctive normal form divide-and-conquer, dominance relation duplicate elimination dynamic programming, dynamic programming - components dynamic programming - space dynamic programming - when applicable edges edit distance element uniqueness exhaustive search algorithm exponential time Federal sentencing guidelines Fibonacci numbers flat earth model Floyd-Warshall algorithm formal language theory forward edge file:///E|/LEC/LECTUR16/INDEX_AL.HTM (3 of 8) [19/1/2003 1:37:56]

Index (complete)

Fourier transform fragment assembly frequency distribution friendship graph gadget Garey and Johnson geometric algorithms, geometric series girth gossip graphs graph traversal greedy heuristic, guessing module Hamiltonian cycle Hamiltonian cycle - hardness Hamiltonian path, harmonic numbers hash function hash table heap data structure heapify heapsort, Heep, Danny heuristic Huffman codes incremental algorithms independent set induction, inorder traversal insert - binary search tree insert - red-black tree insertion sort, instance integer partition integer programming - hardness, file:///E|/LEC/LECTUR16/INDEX_AL.HTM (4 of 8) [19/1/2003 1:37:56]

Index (complete)

inverse exponential function Kruskal's algorithm, linked list local replacement logarithm logic problem longest increasing subsequence longest path lower bound lower bound - sorting mainframe computers Maple master theorem, , Mathematica mathematical induction matrix multiplication, matrix multiplication - applications maze median mergesort minimum spanning tree, mode monotonic sequence morphing motion planning movie stars multigraph nearest neighbor heuristic non-comparison based sorting non-deterministic Turning machine NP NP-completeness NP-completeness - dealing with it open addressing optimization problem optimizing programs for speed order-statistic tree file:///E|/LEC/LECTUR16/INDEX_AL.HTM (5 of 8) [19/1/2003 1:37:56]

Index (complete)

parenthesization partial order partitioning step path path compression permutations - generation pivot pointers polygon triangulation polynomial time predecessor Prim's algorithm principle of optimality priority queue, problem, problem solving techniques, pseudocode P vs. NP, queens problem queue,

,

quicksort, quicksort - analysis RAM model of computation random-access machine randomized algorithm, recurrence relations recursion tree, red-black tree red-black trees

,

reduction, reduction - direction of refrain refrigerator restriction robot arm

file:///E|/LEC/LECTUR16/INDEX_AL.HTM (6 of 8) [19/1/2003 1:37:56]

Index (complete)

Rolling Stones rotation rules for algorithm design satisfiability satisfiability - Cook's theorem selection selection sort, sentence disambiguation sentinel series convergence set data structures, Shifflett shortest path, simple graph simulations social security number solving recurrences songs, complexity of sorting sorting - applications sorting - lower bound, space complexity spanning tree sparse graphs spouse, finding the optimal square of graph stack, , Strassen's algorithm string matching strongly connected components, subgraph isomorphism subset generation subset sum successor supercomputer, sweepline algorithms file:///E|/LEC/LECTUR16/INDEX_AL.HTM (7 of 8) [19/1/2003 1:37:56]

Index (complete)

symmetry, exploiting telephone book telephone keypad telephone network, terrorist text compression topological sorting, transportation problems traveling salesman problem,

,

,

,

traveling salesman problem - approximation, triangulation triangulation - minimum length Turing machine undirected graph union-find data structure upper bound vertex cover, vertex cover - approximation vertex cover - hardness vertex cover - tree vertices Viterbi algorithm worst-case complexity

Algorithms Mon Jun 2 14:49:38 EDT 1997

file:///E|/LEC/LECTUR16/INDEX_AL.HTM (8 of 8) [19/1/2003 1:37:56]

Ranger - Nearest Neighbor Search in Higher Dimensions

Ranger - Nearest Neighbor Search in Higher Dimensions Ranger is a tool for visualizing and experimenting with nearest neighbor and orthogonal range queries in high-dimensional data sets, using multidimensional search trees. It was developed by Michael Murphy as his masters project under Steven Skiena at Stony Brook. Four different search data structures are supported by Ranger: ●

●

●

●

Naive k-d -- the original kd-tree defined by Bentley, "Multidimensional Binary Search Trees Used for Associative Searching" ACM Sept. 1975 Vol. 18. No. 9 Median k-d -- A refined kd-tree, using median cuts and bucketing, discussed in J.H. Friedman, J.L. Bentley, R.A. Finkel "An Algorithm for Finding Best Matches in Logarithmic Expected Time" ACM Transactions of Mathematical Software Vol 3 No. 3 Sept. 1977 pp. 209-226 Sproull k-d -- Sproull's variant of the $kd$-tree. The choice of partition plane is not orthogonal, rather, it is selected by the principal eigenvector of the covariance matrix of the points. R.F. Sproull "Refinements to Nearest-Neighbor Searching in k-Dimensional Trees" J. Algorithmica 1990. pp. 579-589 VP-tree - The vantage point tree is a data structure that chooses vantage points to perform a spherical decomposition of the search space. Yianilos claims this method is suited for nonMinkowski metrics (unlike kd-trees) and for lower dimensional objects embedded in a higher dimensional space. P.N. Yianilos "Data Structures and Algorithms for Nearest Neighbor Search in General Metric Spaces" SODA '93

Ranger supports queries in up to 25 dimensions for all data structures, under any Minkowski metric. Each of these data structures can be applied each of several different applications, including $k$-nearest neighbor graphs and orthogonal range queries. Timing data and algorithm animations can be used to study the performance of these data structures. Because kd-trees are heuristic data structures, their performance can be dramatically effected by the distribution of the data. {\em Ranger} includes generators for a variety of distributions in arbitrary dimensions, including several degenerate distributions Bentley has noted tend to frustrate heuristic data structures. Ranger provides a number of features to aid in visualizing multi-dimensional data. Arbitrary two- or three-dimensional projections of a higher dimensional data set can be viewed. To identify the most appropriate projection at a glance, Ranger provides a d*d matrix of all two-dimensional projections of the data set. In any three-dimensional projections, rotation along arbitrary axes can be used to facilitate understanding.

file:///E|/WEBSITE/IMPLEMEN/RANGER/IMPLEMEN.HTM (1 of 2) [19/1/2003 1:39:44]

Ranger - Nearest Neighbor Search in Higher Dimensions

For each of the four search data structures, a graphic representation of the space division is provided. This provides considerable insight into the appropriateness of a data structure for a particular distribution. Ideally, the search space is decomposed into fat, regularly-sized regions, but this is often not possible with degenerate distributions. Further, Ranger can animate a nearest-neighbor search, by highlighting each region as it is visited, for any two- or three-dimensional projection. Ranger is written in C, using Motif. It runs on Silicon Graphics and HP workstations.

● ●

Link to Michael Murphy's home page Download Files (local site)

Problem Links ● ● ●

Kd-Trees (8) Nearest Neighbor Search (7) Range Search (6)

About the Book Send us Mail Go to Main Page This page last modified on Feb 25, 1996.

file:///E|/WEBSITE/IMPLEMEN/RANGER/IMPLEMEN.HTM (2 of 2) [19/1/2003 1:39:44]

DIMACS Implementation Challenges

DIMACS Implementation Challenges DIMACS is the NSF Science and Technology center for Discrete Mathematics and Theoretical Computer Science, based at Rutgers University. Each year they sponsor an implementation challenge workshop to stimulate research in empirical algorithms research. Each year has a different theme, resulting in implementations for a different class of problems. Proceedings volumes for each workshop are published in the DIMACS series by the American Mathematical Society (AMS). The First DIMACS Implementation Challenge on Network Flows and Matching in October 1991 \cite{JM-92}. Several implementations for maximum weight and maximum cardinality matching were collected and can be obtained by anonymous ftp from dimacs.rutgers.edu in the directory pub/netflow/matching. These include: ●

●

●

●

●

A maximum cardinality matching solver in Fortran 77 by R. Bruce Mattingly %([email protected]) and Nathan P. Ritchey ([email protected]) of the Youngstown State University, which seems capable of solving instances of 5,000 nodes and 60,000 edges in under 30 seconds. A maximum cardinality matching solver in C by Edward Rothberg ([email protected]) of Silicon Graphics, which implements Gabow's \cite{Gabow-76} $O(n^3)$ algorithm. A maximum weighted matching solver in C by Edward Rothberg. This is slower than his unweighted solver described above, but obviously more general. For example, it took over 30 seconds on a weighted graph with 500 nodes and 4,000 edges. A preflow-push network flow implementation in C language by Edward Rothberg of Silicon Graphics ([email protected]). It took under a second on a test graph of 500 nodes and 4,000 edges, but over an hour with 5,000 nodes and 40,000 edges. An implementation of 11 network flow variants in C, including the older Dinic and Karzanov algorithms by Richard Anderson ([email protected]) and Joao Setubal of the University of Washington. On an instance of 8,000 vertices and 12,000 edges, all options finished within two seconds.

Programs for the closely related problems of finding cliques, chromatic number, and independent sets were also sought for the second DIMACS challenge. The programs and data from the challenge are available by anonymous ftp from dimacs.rutgers.edu. Source codes are available under pub/challenge/graph and test data under pub/djs. In particular, two C language programs by David S. Johnson and David L. Applegate are available: ●

●

The dfmax.c implements a simple-minded branch-and-bound algorithm very similar to that of Carraghan and Paradalos. for cliques which should perform well on sparse graphs. The dmclique.c is a variant on the simple `semi-exhaustive greedy' scheme for finding large independent sets used in the graph coloring algorithm XRLF. nmclique.c is a new version of dmclique.c that prints out the best clique found.

file:///E|/WEBSITE/IMPLEMEN/DIMACS/IMPLEMEN.HTM (1 of 3) [19/1/2003 1:39:45]

DIMACS Implementation Challenges

Performance data on both programs is available in files results.dfmax and results.dmclique within the directories /pub/challenge/graph/benchmarks/clique and /pub/challenge/graph/benchmarks/volume. Also included in the Second DIMACS Implementation Challenge was satisfiablity. Programs and data from the challenge are available by anonymous ftp from dimacs.rutgers.edu in the directory /pub/challenge/sat. In particular, there is random formula generator named mwff.c for constructing hard satisfiability instances in C by Bart Selman . The Third DIMACS challenge, held in 1994, focused on parallel algorithms for graphs and game-tree search. The Fourth DIMACS challenge, held in 1995, focused on two problems in computational biology; fragment assembly and sorting with reversals. The Fifth DIMACS implementation challenge in 1996 will focus on elementary data structures like dictionaries. The world's best available dictionary implementations are likely to be identified in the course of the challenge.

● ● ● ● ● ● ●

Link to the DIMACS home page Challenge 1 - network flow and matching Challenge 2 - clique, coloring, and satisfiability Challenge 3 - parallel algorithms Challenge 4 - computational biology; sequence assembly and genome rearrangements Challenge 5 - priority queues, dictionaries, and multi-dimensional point sets. Download Files (local site)

Problem Links ● ● ● ● ● ● ● ● ●

Clique (9) Matching (9) Network Flow (8) Satisfiability (8) Independent Set (7) Vertex Coloring (7) Vertex Cover (5) Dictionaries (1) Kd-Trees (1) file:///E|/WEBSITE/IMPLEMEN/DIMACS/IMPLEMEN.HTM (2 of 3) [19/1/2003 1:39:45]

DIMACS Implementation Challenges ●

Priority Queues (1)

About the Book Send us Mail Go to Main Page This page last modified on Feb 23, 1996.

file:///E|/WEBSITE/IMPLEMEN/DIMACS/IMPLEMEN.HTM (3 of 3) [19/1/2003 1:39:45]

Stony Brook Project Implementations

Stony Brook Project Implementations These files represent implementations produced by my students at Stony Brook in the course of research or course projects. They were not intended for distribution, often have not been tested extensively, and hence should be viewed even more suspiciously than other codes available on this WWW site. However, they are likely to be useful at least (if only) as models, in particular for those problems where I have been unable to identify alternative codes. ●

●

●

●

●

Edge Coloring -- Yan Dong's implementation of Vizing's theorem in C++ provides an edge coloring of any graph using at most one more color than required. Check out the distib/Vizing directory. Suffix Trees (1) -- Dimitris Margaritis' implementation of suffix trees was used in our experiments with interactive sequencing by hybridization. Check out the distrib/dmarg directory. Suffix Trees (2) -- Ting Chen's implementation of suffix trees was used in a fragment assembler for shotgun DNA sequencing. Check out the distrib/tichen directory. Bandwidth (1), (2) -- These are programs for exact solution of the bandwidth of a graph by Ingmar Bitter, and Dario Vlah, as class assignments. Ingmar's is likely the best but also the bulkiest. Check out the distrib/bandwidth directory.

Download Files (local site)

Problem Links ● ● ●

Bandwidth Reduction (6) Edge Coloring (6) Suffix Trees and Arrays (6)

About the Book Send us Mail Go to Main Page This page last modified on Aug 5, 1996.

file:///E|/WEBSITE/IMPLEMEN/STONY/IMPLEMEN.HTM [19/1/2003 1:39:46]

Neural-Networks for Cliques and Coloring

Neural-Networks for Cliques and Coloring This directory contains codes by Arun Jagota ([email protected]) and Laura Sanchis ([email protected]) exploring neural-network approaches to approximate the solution of maxclique and vertex coloring. Implementations in both C and Mathematica are provided, along with associated journal references. ●

Download Files (local site)

Problem Links ● ● ● ●

Clique (6) Vertex Coloring (6) Vertex Cover (6) Independent Set (5)

About the Book Send us Mail Go to Main Page This page last modified on Aug 26, 1996.

file:///E|/WEBSITE/IMPLEMEN/CLIQUE/IMPLEMEN.HTM [19/1/2003 1:39:47]

Clarkson's higher dimensional convex hull code

Clarkson's higher dimensional convex hull code Hull is an ANSI C program by Ken Clarkson of Bell Laboratories that computes the convex hull of a point set in general (but small!) dimension. The input is a list of points, and the output is a list of facets of the convex hull of the points, each facet presented as a list of its vertices. (The facets are assumed to be simplices, such as triangles in 3d; this is enforced by tiebreaking, giving a triangulation of a facet by "placing".) The program can also compute Delaunay triangulations and alpha shapes, and volumes of Voronoi regions. The program uses exact arithmetic when possible, with a moderate speed penalty. (Typically a factor of 2 or 3 for Delaunay triangulation, less for convex hulls). Output in postscript and OFF format for geomview is supported.

● ●

Ken Clarkson's home page Download Files (local site)

Problem Links ●

Convex Hull (6)

About the Book Send us Mail Go to Main Page This page last modified on Jun 2, 1997.

file:///E|/WEBSITE/IMPLEMEN/CLARKSON/IMPLEMEN.HTM [19/1/2003 1:39:47]

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

From [email protected] Wed Apr 2 17:03:32 1997 Received: from cs.brown.edu (cs.brown.edu [128.148.128.2]) by cs.sunysb.edu (8.6.12/8.6.9) with ESMTP id RAA07268 for <[email protected]>; Wed, 2 Apr 1997 17:02:50 -0500 Received: from dumba.cs.brown.edu (dumba.cs.brown.edu [128.148.38.84]) by cs.brown.edu (8.8.5/8.7.1) with ESMTP id QAA20853 for <[email protected]>; Wed, 2 Apr 1997 16:52:50 -0500 (EST) Received: (rt@localhost) by dumba.cs.brown.edu (8.8.3/BrownCS1.0) id QAA14003; Wed, 2 Apr 1997 16:52:48 -0500 (EST) Date: Wed, 2 Apr 1997 16:52:48 -0500 (EST) Message-Id: <[email protected]> From: Roberto Tamassia To: Steve Skiena <[email protected]> Subject: airplane data References: <[email protected]> Status: RO X-Status: Hi Steve, I'm enclosing a database of flights between US cities that I collected with EasySABRE in 1992. I have several variations of a programming assignment that uses the data. I'm enclosing in the next messages the tex and dvi files of the '95 version of the assignment. I am preparing a Java version for this semester's class. It should be ready in about one week. It will be available from the course Web page at http://www.cs.brown.edu/courses/cs016/ I hope this is useful. -Roberto #Airport information - USA # #Fields: # 3 - Airport abbreviation # 1 - <padding> # 5 - Time Zone (offset from GMT) # 1 - <padding> # 3 - X coordinate on map # 1 - <padding> # 3 - Y coordinate on map # 1 - <padding> # remainder - Name of City/Airport # # Route information # Starts after the '!' delimiter # # Fields: # 3 - Source # 1 - <padding> # 3 - Destination # ABQ -0800 195 275 Albuquerque, New Mexico ATL -0600 470 280 Atlanta, Georgia BNA -0700 430 240 Nashville, Tennessee BOS -0600 590 100 Boston, Massachusetts DCA -0600 540 180 Washington D.C. (National) DEN -0800 215 205 Denver, Colorado DFW -0700 310 305 Dallas/Fort Worth, Texas file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (1 of 67) [19/1/2003 1:40:21]

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

DTW HOU JFK LAX MIA MSP MSY ORD PHL PHX PVD RDU SEA SFO STL TPA ! ABQ ABQ ABQ ABQ ATL ATL ATL BNA BNA BOS DCA DCA DEN DEN DEN DFW DFW DTW HOU JFK LAX LAX MIA MSP MSY MSY MSY ORD PVD PHX SEA

-0600 -0700 -0600 -0900 -0600 -0700 -0700 -0700 -0600 -0800 -0600 -0600 -0900 -0900 -0700 -0600

445 330 565 55 535 340 405 410 550 120 595 530 55 10 380 500

140 355 130 270 390 115 345 155 155 290 122 230 45 190 210 360

Detroit, Michigan Houston, Texas (Hobby) New York (Kennedy) Los Angeles, California Miami, Florida Minneapolis/St Paul, Minnesota New Orleans, Louisiana Chicago, Illinois Philadelphia, Pennsilvania/Wilmington, Delaware Phoenix, Arizona Providence/Newport, Rhode Island Raleigh/Durham, North Carolina Seattle/Tacoma, Washington San Francisco, California St Louis, Missouri Tampa, Florida

DEN DFW LAX PHX BNA MSY RDU ORD STL JFK JFK RDU DFW LAX ORD HOU STL ORD MIA PHL PHX SFO TPA ORD DFW HOU TPA STL MIA SFO SFO

#Airline Flight Schedule - USA # #Fields: # 2 - Airline # 4 - Flight # 2 - <padding> # 3 - Origin Airport # 1 - <padding> # 5 - Departure Time # 2 - <padding> # 3 - Destination Airport # 1 - <padding> file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (2 of 67) [19/1/2003 1:40:21]

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

# 5 - Arrival Time # 2 - <padding> # 2 - Meals (S=snack,L=lunch,D=dinner,B=breakfast,#=depends on class) # 4 - <padding> # 1 - Stops during flight # 4 - <padding> # 3 - Aircraft type # 12 - <padding> # remainder - Booking Classes # AA 748 ABQ 257P ATL 1047P S 1 S80 F Y M Q B TW 410 ABQ 628A DCA 132P # 1 M80 F Y B Q M DL 660 ABQ 125P DCA 845P LD 1 757 F Y B M Q UA 785 ABQ 645A DEN 757A S 0 733 F Y B M Q YV 11 ABQ 715A DEN 920A 1 BE1 Y Q L H M YV 221 ABQ 820A DEN 1010A 1 BE1 Y Q L H M UA 982 ABQ 829A DEN 939A S 0 72S F Y B M Q CO1594 ABQ 830A DEN 944A S 0 73S F A Y Q H UA 718 ABQ 1020A DEN 1132A 0 72S F C Y B M CO1708 ABQ 1026A DEN 1139A 0 733 F A Y Q H YV 687 ABQ 1045A DEN 1235P 1 BE1 Y Q L H M UA 470 ABQ 114P DEN 224P 0 72S F Y B M Q CO1660 ABQ 210P DEN 322P 0 733 A Y Q H K YV 237 ABQ 300P DEN 455P 1 BE1 Y Q L H M UA 671 ABQ 325P DEN 437P 0 733 F Y B M Q UA 716 ABQ 449P DEN 604P 0 733 F Y B M Q YV 607 ABQ 500P DEN 650P 1 BE1 Y Q L H M CO1184 ABQ 525P DEN 639P # 0 73S A Y Q H K AA 202 ABQ 450A DFW 721A 0 72S FN YN B M H DL1042 ABQ 630A DFW 901A # 0 72S FN YN BN MN QN AA 880 ABQ 915A DFW 1151A V 0 72S F Y B M H DL 768 ABQ 945A DFW 1223P S 0 M80 F Y B M Q AA1050 ABQ 1053A DFW 134P L 0 72S F Y B M H CO 829 ABQ 1120A DFW 358P L 1 73S A Y Q H K DL 660 ABQ 125P DFW 403P L 0 757 F Y B M Q AA1480 ABQ 140P DFW 418P 0 S80 F Y B M H YV 84 ABQ 200P DFW 637P 2 BE1 Y Q L H M AA 592 ABQ 409P DFW 701P D 0 S80 F Y B M H DL1048 ABQ 455P DFW 730P D 0 72S F Y B M Q AA1044 ABQ 600P DFW 832P D 0 72S F Y B M H DL1096 ABQ 825P DFW 1054P 0 M80 FN YN BN MN QN WN 515 ABQ 610A HOU 905A 0 733 Y K L B Q WN 29 ABQ 1005A HOU 225P 2 733 Y K L B Q WN 31 ABQ 1050A HOU 255P 1 73S Y K L B Q WN 803 ABQ 1105A HOU 330P 2 73S Y K L B Q WN 539 ABQ 845P HOU 1135P 0 73S Y K L B Q WN 530 ABQ 700A LAX 750A 0 73S Y K L B Q WN 532 ABQ 950A LAX 1045A 0 73S Y K L B Q WN 343 ABQ 1125A LAX 1225P 0 73S Y K L B Q WN 994 ABQ 100P LAX 250P 1 73S Y K L B Q WN 430 ABQ 410P LAX 505P 0 73S Y K L B Q WN 931 ABQ 510P LAX 700P 1 73S Y K L B Q WN 362 ABQ 620P LAX 715P 0 733 Y K L B Q WN 590 ABQ 730P LAX 825P 0 733 Y K L B Q AA 880 ABQ 915A MIA 439P VL 1 72S F Y M Q B UA 982 ABQ 829A MSP 124P SL 1 72S F Y B M Q DL1042 ABQ 630A MSY 1105A # 1 72S F@ Y@ B@ M@ Q@ WN 41 ABQ 250P MSY 700P 1 73S Y K L B Q AA1186 ABQ 845A ORD 1227P B 0 S80 F Y B H M AA 134 ABQ 1154A ORD 339P L 0 S80 F Y B H M

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (3 of 67) [19/1/2003 1:40:21]

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AA 748 TW 88 AA1044 WN 706 HP 650 WN 600 WN 483 HP 653 WN 386 WN 866 HP 836 WN 994 WN 552 HP 845 WN1745 WN 931 HP 647 WN 563 WN 812 HP1221 WN 814 WN 416 HP 845 WN 592 WN 965 WN 971 WN 515 TW 410 TW 844 WN 535 TW 290 UA 470 TW 88 DL 545 AA1271 DL1166 DL1247 DL1045 DL1709 DL 376 DL 657 DL 735 DL 850 DL 202 NW1856 DL 350 DL 410 UA1032 NW1866 DL1610 DL 378 DL 356 DL1420 DL 326 DL1406 DL1258 DL 314 US1642 NW1856 DL 710 DL1436

ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL

257P 328P 600P 650A 700A 745A 830A 1000A 1100A 1145A 1230P 100P 235P 255P 410P 510P 530P 620P 725P 745P 910P 1000P 255P 820A 205P 510P 610A 628A 844A 1115A 1244P 114P 328P 1144A 440P 632A 1000A 1154A 327P 507P 708P 1147P 624A 636A 750A 902A 1003A 1003A 1240P 119P 319P 330P 510P 705P 859P 1157P 625A 650A 750A 819A 1000A

ORD ORD ORD PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX SEA SFO SFO SFO STL STL STL STL STL STL STL ABQ ABQ BNA BNA BNA BNA BNA BNA BNA BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS DCA DCA DCA DCA DCA

645P 856P 1204A 805A 814A 855A 940A 1115A 1210P 100P 145P 210P 345P 410P 520P 625P 645P 730P 835P 859P 1015P 1105P 718P 950A 325P 635P 1120A 950A 1203P 220P 403P 601P 651P 405P 826P 625A 1000A 1150A 330P 510P 710P 1140P 845A 1004A 1121A 1125A 1220P 137P 426P 350P 655P 552P 740P 935P 1125P 210A 755A 1059A 924A 951A 1140A

S # D

S

B B # S D L D #

B B BS B B S LS L S S D D # B S B B S

0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 1 0 2 1 0 0 0 0 0 0 0 0 1 1 0 0 1 1 0 1 0 0 0 0 0 0 2 0 0 0

S80 M80 72S 733 733 73S 733 733 73S 733 733 73S 73S 73S 733 73S 733 733 733 733 733 73S 73S 733 73S 733 733 M80 L10 73S 72S 72S M80 72S 72S 73S D9S D9S 757 M80 M80 D9S 767 72S D9S 72S 757 72S D9S 763 72S 72S L10 757 757 763 757 D9S D9S 757 757

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (4 of 67) [19/1/2003 1:40:21]

F F F Y Y Y Y F Y Y Y Y Y Y Y Y F Y Y F Y Y Y Y Y Y Y F F Y F F F F F FN F F F F F FN FN F@ F F F F F F F F F F F FN FN FN FN F F

Y Y Y K B K K Y K K B K K B K K Y K K Y K K B K K K K Y C K Y Y Y Y Y YN Y Y Y Y Y YN YN Y@ Y Y Y Y Y Y Y Y Y Y Y YN YN YN YN Y Y

B B M L H L L B L L H L L H L L B L L B L L H L L L L B Y L B B B B H BN B B B B B BN BN B@ B B B B B B B B B B B BN BN BN BN B B

H Q Q B K B B H B B K B B Q B B H B B H B B Q B B B B Q B B Q M Q M B MN M M M M M MN MN M@ M M M M M M M M M M M MN MN HN M M M

M M B Q Q Q Q K Q Q Q Q Q V Q Q Q Q Q K Q Q V Q Q Q Q M Q Q M Q M Q M QN Q Q Q Q Q QN QN Q@ H Q Q Q H Q Q Q Q Q Q QN QN QN H Q Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

DL 464 NW1866 DL 402 DL 378 DL 280 NW1878 DL 178 DL 394 UA 717 CO 275 DL 445 DL 257 UA 373 DL1668 CO1759 DL1055 DL 675 DL 785 AA 837 DL 793 DL 444 DL 551 DL 409 DL 671 DL 803 DL 357 DL 239 DL 775 AA1683 DL 873 DL 17 DL1101 DL 307 DL 545 DL 323 DL 125 DL1428 DL1137 DL1045 DL1457 DL 946 DL 299 DL 837 AA 139 DL 758 DL 179 TW 621 AA 425 DL 145 DL 719 DL1033 DL 557 DL 139 DL 847 DL 691 DL1709 DL 845 AA 385 DL 195 DL1023

ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL

1145A 1240P 120P 319P 449P 545P 658P 846P 640A 650A 857A 1159A 310P 312P 400P 652P 709P 625A 725A 819A 819A 819A 820A 822A 825A 826A 833A 838A 840A 907A 1006A 1144A 1144A 1144A 1144A 1147A 1149A 1154A 1154A 1155A 1157A 1159A 1203P 1225P 122P 124P 139P 156P 310P 310P 310P 312P 314P 323P 325P 327P 327P 430P 449P 455P

DCA DCA DCA DCA DCA DCA DCA DCA DEN DEN DEN DEN DEN DEN DEN DEN DEN DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW

125P 216P 259P 455P 630P 731P 835P 1025P 804A 810A 1005A 105P 429P 415P 520P 1130P 830P 735A 835A 1055A 1109A 1110A 1108A 927A 1055A 1108A 1105A 1108A 1002A 1108A 1110A 1253P 227P 230P 230P 225P 222P 225P 204P 217P 217P 215P 217P 143P 414P 230P 501P 311P 422P 606P 610P 611P 553P 555P 611P 601P 552P 547P 605P 742P

L L L D D D B B B L S S D D D B B # S S # B S B S S B S L S L L L L L L S S S L S L

S S S S

S S D D

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 1 2 2 2 0 1 1 1 1 0 1 0 0 1 1 2 1 1 1 1 1 1 1 1 0 2 0 1 0 0 1 2 2 1 1 2 1 1 0 0 1

757 D9S 757 72S 757 D9S 757 M80 737 733 757 72S 737 72S 733 M80 757 757 72S 763 72S M80 M80 72S 73S M80 72S 73S S80 M80 L15 M80 767 72S M80 73S 72S M80 D9S D9S D9S 73S M80 72S M80 L10 D9S S80 767 M80 M80 757 757 M80 72S 757 73S S80 L10 M80

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (5 of 67) [19/1/2003 1:40:22]

F Y B M Q F Y B M H F Y B M Q F Y B M Q F Y B M Q F Y B M H F Y B M Q F Y B M Q F Y B M Q Y Q H K B F Y B M Q F Y B M Q F Y B M Q F Y B M Q A Y Q H K F@ Y@ B@ M@ Q@ F Y B M Q FN YN BN MN QN FN YN B H M F Y B M Q Y B M Q H F Y B M Q F Y B M Q F Y B M Q F Y B M Q F Y B M Q F Y B M Q F Y B M Q F Y B H M F Y B M Q F Y B M Q F Y B M Q F Y B M H F Y B M Q F Y B M Q F Y B M Q F Y B M Q F Y B M Q F Y B M Q F Y B M Q F Y B M Q F Y B M Q F Y B M Q F Y B H M F Y B M Q F Y B M Q F Y B Q M F Y B H M F Y B M Q F Y B M Q F Y B M Q F Y B M Q F Y B M H F Y B M Q F Y B M Q F Y B M Q F Y B M Q F Y B H M F Y B M Q F Y B M Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

DL 405 DL1153 DL 817 DL 397 DL1055 DL1002 DL1081 DL 883 AA 701 DL1507 DL 593 DL 657 DL 970 DL 227 DL 585 DL 955 NW 492 DL 868 NW 494 DL 964 NW 496 DL1044 NW 490 DL 412 DL 766 NW 498 DL 256 DL 714 NW 823 DL1050 DL 946 DL1170 DL1023 DL 30 DL1916 TW 816 HP 44 DL1565 DL 369 DL 269 DL 179 DL 139 DL 157 DL1049 DL 187 DL 446 DL 143 DL 339 DL 321 AA 709 PA 425 DL1155 AA1555 DL 449 AA 825 DL 395 DL 472 DL 491 AA 471 DL 325 DL 469

ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL

641P 641P 642P 649P 652P 653P 656P 656P 700P 702P 703P 708P 743P 835P 1015P 1145P 640A 826A 950A 1004A 1250P 127P 400P 448P 505P 555P 859P 1147P 820A 958A 1157A 119P 455P 1230P 145P 234P 815A 823A 828A 1200N 124P 314P 329P 457P 707P 851P 1035P 632A 819A 828A 835A 1001A 1100A 1144A 135P 139P 326P 659P 730P 857P 1022P

DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW HOU HOU HOU HOU HOU JFK JFK JFK LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA

755P 840P 940P 936P 937P 937P 923P 925P 832P 946P 917P 937P 939P 948P 1117P 1245A 829A 1010A 1133A 1150A 234P 315P 549P 955P 655P 745P 1050P 125A 1057A 1100A 403P 225P 605P 238P 353P 450P 1125A 955A 1130A 130P 440P 920P 455P 630P 855P 1040P 1210A 810A 1000A 1015A 1019A 1145A 1243P 125P 333P 325P 515P 845P 926P 1035P 1156P

D S D D # D D D S S S

B B

L L S S D D

S S S L D L L S B B BS L LS D D D D S S B B B #

L S D D

0 1 2 1 1 2 1 1 0 1 1 1 1 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 1 0 2 0 0 0 0 0 1 0 1 0 1 3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

M80 D9S M80 M80 M80 M80 D9S 73S S80 757 D9S M80 73S 763 M80 72S DC9 72S 72S M80 D9S M80 D9S 757 757 D9S 763 M80 D9S D9S D9S M80 M80 L15 310 72S 320 L10 757 L10 L10 757 L10 L10 767 757 757 757 757 727 72S 763 72S 763 72S 72S 757 767 72S 72S 72S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (6 of 67) [19/1/2003 1:40:22]

F F@ F@ F@ F@ F@ F@ F@ F F F@ F F@ FN FN FN FN F FN F F F F F F F F FN F F F F F F F F F F F F F F F F F F FN FN F F F F F F F F F F F F FN

Y Y@ Y@ Y@ Y@ Y@ Y@ Y@ YN Y Y@ Y Y@ YN YN YN YN Y YN Y Y Y Y Y Y Y Y YN Y Y Y Y Y C C C Y Y Y Y Y Y Y Y Y Y YN YN Y Y Y Y Y Y Y Y Y Y Y Y YN

B B@ B@ B@ B@ B@ B@ B@ B B B@ B B@ BN BN BN BN B BN B B B B B B B B BN B B B B B Y Y Y B B B B B B B B B B BN BN B B B B B

M M@ M@ M@ M@ M@ M@ M@ H M M@ M M@ MN MN MN M M M M M M M M

Q@ Q@ Q@ Q@ Q@ Q@ Q@ M Q Q@ Q Q@ QN QN QN H Q H Q H Q H Q

M H M Q MN QN M H M Q M Q M Q M Q B M B M B Q H K M Q M Q M Q M Q M H M Q M Q M Q M Q MN QN MN QN M M H M Q M H M H

B M H B M H B M H B M Q B M H B M Q BN MN QN

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

NW 673 DL 954 NW 759 NW 763 DL1085 DL1855 NW1027 DL 700 DL 793 DL 307 DL 139 DL1199 DL1507 DL 577 DL 991 DL 596 AA 833 UA 507 DL1064 AA1221 DL 820 UA 721 DL 598 AA 357 DL 142 UA 631 DL1124 UA 944 AA1271 DL 692 CO 235 UA 415 DL 335 AA 673 UA 633 DL 298 DL 574 DL 202 US1238 DL 726 US1714 DL 686 UA1236 DL 860 US 274 DL 842 US 836 DL1118 CO 543 DL 444 DL 369 HP 838 DL 295 DL1855 HP3003 DL 901 US1238 UA1456 US1202 CO1016

ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL

710A 833A 900A 1200N 1202P 448P 510P 702P 819A 1144A 314P 457P 702P 856P 1032P 626A 715A 755A 821A 1010A 1014A 1120A 1147A 105P 121P 140P 333P 359P 440P 446P 500P 512P 656P 757P 827P 837P 1145P 636A 715A 1018A 1110A 119P 125P 447P 450P 701P 830P 844P 730A 819A 828A 900A 1146A 448P 455P 657P 715A 730A 925A 130P

MSP MSP MSP MSP MSP MSP MSP MSP MSY MSY MSY MSY MSY MSY MSY ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHX PHX PHX PHX PHX PHX PHX PHX PVD PVD PVD PVD

843A 1010A 1032A 135P 145P 615P 650P 840P 850A 1216P 345P 525P 735P 930P 1050P 715A 804A 852A 920A 1100A 1105A 1213P 1235P 155P 220P 234P 430P 815P 534P 550P 815P 614P 750P 845P 921P 940P 1230A 825A 906A 1215P 1258P 315P 455P 640P 638P 858P 1030P 1040P 1100A 110P 1025A 1110A 135P 1050P 655P 900P 1057A 1100A 1259P 531P

B B B L L D D D # S S S

B B B B

S L # L L # # D D # D D # # B B S L # D S D S B SL B S L D D D B B S S

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 3 0 0 0 2 0 0 1 1 1 1

DC9 M80 D9S D9S 73S 72S M80 73S 763 767 757 M80 757 767 72S 72S S80 73S 72S 727 72S 72S 72S S80 757 72S 72S 733 72S 767 737 72S 72S S80 72S 757 72S 72S 73S M80 D9S 757 73S 763 D9S M80 100 M80 72S 72S 757 733 757 72S 320 757 73S 73S 733 737

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (7 of 67) [19/1/2003 1:40:22]

FN F FN F F F F F F F F Y F F FN FN F F F F F F F F F F F F F F A F F F F F FN FN FN F F F F F Y F F F Y Y F Y F F@ F F FN F F A

YN Y YN Y Y Y Y Y Y Y Y B Y Y YN YN Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y YN YN YN Y Y Y Y Y B Y Y Y Q B Y B Y Y@ Y Y YN Y Y Y

BN B BN B B B B B B B B

M M M M M M M M M M M

H Q H H Q Q H Q Q Q Q

B B BN BN B B B B B B B B B B B B B B Q B B B B B BN BN BN B B B B B

M M MN MN M M M M M M M M M M M M M M H M M M M M MN MN HN M H M M M

Q Q QN QN H Q Q H Q Q Q H Q Q Q Q H Q K H Q H Q Q QN QN QN Q Q Q Q Q

B M Q B H Q B M Q H K B M Q H B M Q H K Q B M Q B@ M@ Q@ B H K B M Q BN HN QN B M Q B H Q Q H K

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

US1518 DL1458 NW 498 DL 470 DL 806 US1642 AA 190 DL1100 AA1006 DL 548 US1633 DL 412 AA 868 DL 724 DL1136 DL1202 DL 835 DL1709 DL 197 AA 673 DL 625 DL 270 DL 307 DL 977 DL 382 DL 691 DL1411 DL 143 TW 73 DL 696 TW 659 DL1260 TW 621 TW 599 DL1435 DL 878 TW 609 DL1057 DL 273 DL1175 DL 539 DL 790 DL1597 DL 317 DL1003 DL 765 DL 389 DL 474 DL 885 DL1154 DL 947 DL 784 DL 511 DL1167 AA 132 AA 566 AA 530 AA 350 AA1574 US1082 AA1334

ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA

140P 149P 555P 645P 630A 650A 702A 1011A 115P 144P 215P 448P 540P 658P 847P 1143P 926A 327P 646P 757P 1031P 841A 1144A 1209P 321P 325P 719P 1035P 742A 1016A 1039A 137P 139P 439P 505P 701P 756P 1026P 659A 833A 1154A 141P 330P 457P 702P 859P 1027P 1155P 535A 710A 1040A 1230P 350P 725P 830A 155P 735P 910A 215P 705P 730P

PVD PVD PVD PVD RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU SEA SEA SEA SEA SEA SFO SFO SFO SFO SFO SFO SFO STL STL STL STL STL STL STL STL STL STL TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA ATL ATL ATL ATL ATL ATL BOS BOS BOS DCA DCA DCA DCA

508P 410P 959P 900P 740A 926A 815A 1130A 226P 300P 438P 600P 657P 815P 1010P 1255A 1130A 900P 905P 1207A 130A 1045A 450P 215P 530P 855P 920P 159A 827A 1050A 1127A 220P 224P 527P 540P 740P 840P 1055P 815A 1000A 120P 305P 450P 620P 830P 1027P 1145P 110A 732A 908A 1237P 228P 544P 920P 1147A 519P 1054P 1146A 456P 1047P 1002P

S S D D # S

S

S S S

B SD D # S B S L D D D S B

# # # S # # S

S S

# #

B S D

# D

1 0 1 0 0 1 0 0 0 0 1 0 0 0 0 0 0 2 0 1 1 0 2 0 0 3 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0

D9S M80 D9S M80 72S D9S 727 73S 727 72S D9S 757 S80 73S 72S 73S L10 757 763 S80 72S 767 767 757 763 72S L10 757 D9S D9S D9S 73S D9S D9S 72S D9S D9S D9S 763 72S 757 763 72S 757 763 757 763 767 D9S 73S D9S M80 72S D9S S80 S80 S80 757 72S 100 S80

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (8 of 67) [19/1/2003 1:40:22]

F F F F FN FN F F F F F F F F F FN F F F F FN F F Y F F F FN F F F F F F F F F FN FN F F F F F F F FN FN FN F F F F F F F F F F F F

Y Y Y Y YN YN Y Y Y Y Y Y Y Y Y YN Y Y Y Y YN Y Y B Y Y Y YN Y Y Y Y Y Y Y Y Y YN YN Y Y Y Y Y Y Y YN YN YN Y Y Y Y Y Y Y Y Y Y Y Y

B B B B BN BN B B B B B B B B B BN B B B H BN B B M B B B BN B B B B B B B B B BN BN B B B B B B B BN BN BN B B B B B B B B B B B B

H M M M MN HN H M H M H M H M M MN M M M B MN M M Q M M M MN Q M Q M Q Q M M Q MN MN M M M M

Q Q H Q QN QN M Q M Q M Q M Q Q QN Q Q Q M QN Q H H Q Q Q QN M Q M Q M M Q Q M QN QN Q Q Q Q

M Q M Q MN QN MN QN MN QN M Q M Q M Q M Q M Q H M H M H M H M H M H Q H M

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AA1615 AA1543 AA1339 AA 199 DL1087 AA1319 DL1045 AA1337 AA 557 AA1227 DL1709 AA 521 DL 657 AA1621 NW1102 WN 436 AA1212 NW1444 NW1524 AA 604 NW1150 WN1615 AA 860 NW1582 WN 443 WN 741 WN 485 WN 413 WN 758 WN 455 TW 792 WN 440 AA1045 AA1661 AA1517 AA1651 AA 687 AA 663 NW1051 NW 709 WN 443 AA1125 AA 881 WN 758 AA1501 AA 811 UA1269 AA 188 AA1285 UA 511 AA 628 AA1475 UA 365 AA 593 AA 283 AA1041 US 138 AA1330 AA 780 US 843

BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA

1000A 854P 640A 810A 855A 1000A 1220P 100P 210P 335P 405P 701P 745P 844P 710A 730A 840A 1015A 130P 200P 500P 605P 727P 750P 725A 825A 1235P 500P 720P 955P 1221P 840A 950A 315P 904P 840A 205P 740P 835A 420P 725A 945A 320P 720P 904P 643A 728A 830A 947A 1010A 157P 308P 420P 530P 738P 854P 630A 840A 205P 400P

DEN DEN DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW HOU HOU HOU HOU HOU HOU JFK LAX LAX LAX LAX MIA MIA MIA MSP MSP MSY MSY MSY MSY MSY ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD PHL PHL PHL PHL

1145A 1034P 838A 1008A 1045A 249P 204P 308P 423P 544P 601P 908P 937P 1037P 937A 1115A 1105A 1243P 401P 430P 729P 935P 955P 1016P 1045A 1020A 230P 655P 1035P 1140P 450P 1200N 1216P 541P 1136P 1148A 525P 1051P 1049A 634P 925A 1113A 448P 920P 1035P 821A 858A 1005A 1115A 1139A 334P 444P 600P 716P 916P 1035P 915A 1126A 456P 645P

S B B S L L

S D S S V S

S D

S L D S B D # S

V S V

D D B B S

0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 2 0 0 0 2 0 1 1 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

S80 S80 757 757 72S 727 D9S S80 S80 S80 757 S80 M80 72S D9S 73S S80 D9S D9S S80 D9S 733 S80 DC9 733 735 73S 73S 73S 733 72S 733 757 757 757 72S 72S 72S D9S D9S 733 S80 72S 73S S80 100 72S 100 S80 72S S80 72S 73S S80 S80 100 D9S S80 S80 D9S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (9 of 67) [19/1/2003 1:40:22]

F F F F F F F F F F F F F F FN Y F F F F F Y F FN Y Y Y Y Y Y F Y F F FN F F F F F Y F F Y FN FN F F F F F F F F F F FN F F F

Y B H M Y B H M Y B M H Y B M H Y B M Q Y M B Q Y B M Q Y B M H Y B M H Y B M H Y B M Q Y B M H Y B M Q Y B M H YN BN M H K L B Q Y B H M Y B M H Y B M H Y B H M Y B M H K L B Q Y B H M YN BN M H K L B Q K L B Q K L B Q K L B Q K L B Q K L B Q Y B Q M K L B Q Y B H M Y B H M YN B H M Y B H M Y B H M Y B H M Y B M H Y B M H K L B Q Y B M H Y B M H K L B Q Y B M H YN B M H Y B M Q Y B M H Y B M H Y B M Q Y B M H Y B M H Y B M Q Y B M H Y B M H Y B M H YN BN HN QN Y B H M Y B H M Y B H Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AA1222 WN 440 AA 651 AA1691 WN 394 NW1444 AA1128 AA1358 AA1068 AA1339 DL1709 AA1663 AA1287 TW 628 AA1357 TW 529 AA1323 TW 78 TW 717 AA1655 AA1643 AA1645 AA 315 US 599 DL 225 DL 655 DL 205 DL 755 NW1863 DL1149 DL 867 DL 701 DL 263 DL 599 AA1125 AA1691 US1141 AA 749 NW1807 US1507 US1429 NW1853 DL 535 CO 309 US 890 NW1855 TW 391 US 768 NW1857 US 779 NW1859 NW1805 US 187 DL1861 NW1863 US 363 CO 319 NW1865 US1635 NW1867 US 143

BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS

740P 840A 950A 320P 335P 1015A 830A 205P 740P 640A 405P 955A 310P 700A 950A 1009A 310P 341P 732P 859P 840A 205P 740P 310P 630A 820A 945A 1145A 100P 110P 325P 345P 645P 830P 705A 1240P 140P 617P 700A 700A 800A 800A 820A 830A 900A 900A 920A 1000A 1000A 1100A 1100A 1200N 1200N 1240P 100P 100P 130P 200P 200P 300P 300P

PHL PHX PHX PHX PHX PVD RDU RDU RDU SEA SEA SFO SFO STL STL STL STL STL STL STL TPA TPA TPA ABQ ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL BNA BNA BNA BNA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA

1032P 1120A 1230P 602P 605P 309P 1053A 440P 1006P 1146A 900P 1250P 559P 810A 1102A 1122A 414P 451P 840P 1007P 1122A 455P 1024P 751P 907A 1059A 1223P 226P 500P 345P 605P 748P 927P 1105P 905A 236P 446P 812P 834A 841A 937A 946A 1000A 1212P 1039A 1047A 1230P 1137A 1142A 1237P 1240P 141P 139P 215P 244P 237P 459P 344P 339P 444P 439P

D L D S B S BR SD L D #

B D SD B B B L L L S D D S B L S D B B B B B S B B S S S L L L L L L S S S S

0 0 0 0 0 1 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 1 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0

S80 733 S80 S80 733 D9S S80 S80 727 757 757 S80 S80 D9S 72S 727 72S D9S M80 727 72S 72S 72S 733 757 L10 767 757 D9S M80 763 757 72S 763 S80 S80 D9S S80 D9S 733 D9S M80 72S M80 733 757 M80 73S D9S D9S 72S D9S 734 72S D9S D9S 733 M80 733 320 733

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (10 of 67) [19/1/2003 1:40:22]

F Y B H M Y K L B Q F Y H B M F Y H B M Y K L B Q F Y B M H F Y B M H F Y B M H F Y B M H F Y B H M F Y B M Q F Y B H M F Y B H M F Y B Q M F Y B H M F Y B Q M F Y B H M F Y B Q M F Y B Q M F Y B H M F Y B M H F Y B M H F Y B M H F Y B H Q F Y B M Q F Y B M Q F Y B M Q F Y B M Q F Y B M H F Y B M Q F Y B M Q F Y B F Y B M Q FN YN BN MN QN F Y B M H F Y B M H F Y B H Q F Y B M H FN YN BN M H FN YN BN HN QN F Y B H Q F Y B M H F Y B M Q F A Y Q H F Y B H Q F Y B M H F Y B Q M F Y B H Q F Y B M H F Y B H Q F Y B M H F Y B M H F Y B H Q F Y B M Q F Y B M H F Y B H Q A Y Q H K F Y B M H F Y B H Q F Y B M H F Y B H Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

DL 701 NW1869 US1159 US1612 NW1871 NW1873 US1610 NW1875 US1621 NW1877 US1227 DL 663 AA 379 US1402 UA 661 CO1239 UA 699 CO1603 TW 743 UA 243 CO 261 CO 497 UA 861 AA 333 DL1283 AA1371 DL 169 AA 565 AA 607 DL 841 AA 287 DL 487 NW 575 NW 383 NW 745 NW 583 UA 699 NW 49 NW 486 NW1197 TW 15 TW 391 DL1865 TW 743 NW 35 TW 223 CO 459 AA 333 UA 101 UA 897 NW 303 AA 11 US 890 TW 31 DL 891 US 187 NW 43 AA 103 QF 308 UA 133

BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS

345P 400P 400P 500P 500P 600P 600P 700P 700P 800P 800P 845P 615A 720A 810A 830A 225P 245P 335P 340P 445P 500P 516P 644A 815A 830A 1105A 1148A 210P 300P 535P 650P 625A 725A 910A 1215P 225P 330P 525P 640P 617A 920A 245P 335P 400P 600A 640A 644A 700A 805A 815A 900A 900A 904A 940A 1200N 500P 515P 515P 535P

DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DFW DFW DFW DFW DFW DFW DFW DFW DFW DTW DTW DTW DTW DTW DTW DTW DTW JFK JFK JFK JFK JFK LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX

516P 549P 537P 637P 648P 745P 739P 842P 839P 940P 939P 1017P 1037A 1146A 1054A 1120A 755P 530P 839P 744P 846P 750P 755P 959A 1110A 1201P 159P 315P 546P 608P 854P 938P 835A 938A 1127A 230P 609P 550P 742P 855P 725A 1035A 413P 448P 517P 130P 1135A 1219P 1222P 1133A 128P 1218P 316P 217P 155P 608P 833P 832P 832P 904P

S S D D D D S D S S BR BL B B D S D SD SD D D B B B L L S S D D B B # S # # #

B # BL BL BS BL R BL BL SL LD DS D D D

0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0 2 0 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 2 1 1 1 0 1 0 2 1 1 2 0 0 0 0

757 D9S D9S 73S D9S 320 734 D9S 733 M80 733 M80 757 72S 757 M80 733 M80 M80 D10 72S M80 757 D10 757 S80 757 S80 S80 M80 D10 757 757 D9S D10 757 733 D10 72S 72S L10 M80 72S M80 D9S M80 733 D10 D10 767 D10 D10 733 L10 757 734 757 D10 D10 767

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (11 of 67) [19/1/2003 1:40:22]

F Y B M Q F Y B M H F Y B H Q F Y B H Q F Y B M H F Y B M H F Y B H Q F Y B M H F Y B H Q F Y B M H FN YN BN HN QN F Y B M Q F Y B H M F Y B M F Y B M Q A Y Q H K F Y B M Q A Y Q H K F Y B Q M F Y B M Q Y H K B V A Y Q H K F Y B M Q F Y B H M F Y B M Q F Y B H M F Y B M Q F Y B H M F Y B H M F Y B M Q F Y B H M F Y B M Q FN YN BN M H FN YN BN M H F Y B M H F Y B M H F Y B M Q F C Y B M F Y B M H F Y B M H F C Y B Q F Y B Q M F Y B M Q F Y B Q M F C Y B M F Y B Q M A Y Q H K F Y B H M F Y B M Q F C Y B M F Y B M H P Y B H M F Y B M F C Y B Q F Y B M Q F Y B H Q F Y B M H F Y B H M F J Y F Y B M Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AA1245 PA 435 US 433 DL 433 US 781 US 768 US 724 AA1461 PA 429 DL 367 NW 779 US1588 DL 247 AA1577 CO 381 NW 193 NW 303 NW 745 NW 723 NW 725 NW 47 NW 985 NW 361 US 233 AA1125 UA 781 AA 379 UA 101 AA 525 UA 437 NW 3 AA 829 UA 745 AA 275 UA 155 UA 345 AA 149 CO 393 UA 243 AA 267 NW 985 UA 215 AA1361 UA 309 US1723 US 233 US 433 DL1195 US 749 US 595 US 472 US 500 US1141 DL 367 US1631 US1616 US1639 US 373 HP 82 AA1691 HP 76

BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS

700A 700A 740A 825A 920A 1000A 1140A 1220P 220P 220P 430P 500P 510P 610P 630P 635A 815A 910A 1125A 250P 425P 430P 725P 700A 705A 600A 615A 700A 742A 840A 900A 910A 1003A 1205P 1240P 243P 250P 330P 340P 340P 430P 638P 645P 752P 620A 700A 740A 820A 850A 1005A 1140A 1245P 140P 220P 355P 510P 640P 815P 735A 1240P 500P

MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MSP MSP MSP MSP MSP MSP MSP MSP MSY MSY ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHX PHX PHX

1016A 1025A 1241P 1136A 216P 425P 446P 341P 544P 920P 917P 1000P 822P 930P 1215A 850A 1028A 133P 135P 508P 645P 825P 939P 1122A 1113A 743A 754A 844A 923A 1028A 1053A 1043A 1145A 146P 217P 415P 434P 714P 515P 529P 625P 818P 835P 931P 741A 823A 901A 940A 1012A 1125A 100P 206P 300P 337P 516P 631P 805P 936P 1129A 602P 858P

R B SB B # SL LS L S D D D D D # B B # L S D D # SB B B B B B B B B S L L S D S S D D D S S S S # S # #

# B LD D

0 0 1 0 1 2 1 0 0 2 1 1 0 0 2 0 0 1 0 0 0 1 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0

AB3 72S 734 72S 733 73S 734 72S 72S 72S 72S 734 72S 72S 733 320 D10 D10 757 757 D10 72S 757 73S S80 733 757 D10 S80 733 D9S 757 757 D10 D10 72S S80 72S D10 S80 72S D10 S80 72S 734 73S 734 M80 D9S D9S 733 733 D9S 72S 734 734 733 733 757 S80 757

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (12 of 67) [19/1/2003 1:40:22]

F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B A Y Q FN YN BN F Y B F Y B F Y B F Y B F C Y F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B A Y Q F Y B F Y B F Y B F Y B FN YN B FN YN B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B

H M H M M H H H M M M H M H H M M M M M B M M M H M H M H M M H M H M M H H M H M M H M H H H M H H H H H M H H H H H H H

M Q Q Q Q Q M Q Q H Q Q M K H H H H H M H H M Q M Q M Q H M Q M Q Q M K Q M H Q M Q Q Q Q Q Q Q Q Q Q Q M Q Q Q K M K

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AA 563 US 323 AA 223 US 500 US1591 AA 691 NW 575 DL1283 CO1239 TW 53 NW 583 UA 155 NW 47 DL1775 TW 223 DL 347 NW1229 UA 53 US 890 US 187 TW 71 DL 957 UA 861 UA 93 NW 187 TW 61 DL 487 NW 361 TW 223 TW 31 TW 391 TW 53 TW 71 TW 255 DL 175 NW1297 US1041 US 768 DL 863 DL 603 US1588 NW 486 DL 527 TW 123 DL 989 NW1887 DL 539 DL1169 DL 139 DL 915 NW1863 DL 463 DL 701 NW1883 DL 991 DL1243 AA 839 AA1661 AA1105 DL 528

BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA

704A 1115A 1124A 1245P 415P 600P 625A 815A 830A 1203P 1215P 1240P 425P 510P 600A 600A 700A 800A 900A 1200N 320P 335P 516P 530P 530P 630P 650P 725P 600A 904A 920A 1203P 320P 602P 800A 830A 845A 1000A 1215P 325P 500P 525P 850A 943A 730A 900A 905A 1050A 1240P 220P 315P 400P 559P 725P 755P 925P 808A 129P 714P 650A

RDU RDU RDU RDU RDU RDU SEA SEA SEA SEA SEA SEA SEA SEA SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO STL STL STL STL STL STL TPA TPA TPA TPA TPA TPA TPA TPA ABQ ABQ ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL BNA BNA BNA BOS

916A 313P 130P 421P 610P 813P 1126A 335P 205P 557P 514P 551P 925P 1025P 1139A 1110A 1151A 1127A 513P 816P 846P 845P 1022P 914P 857P 1010P 1159P 1227A 822A 1117A 237P 223P 534P 818P 1208P 1155A 1210P 242P 318P 639P 823P 1128P 1230P 140P 915A 1043A 1055A 1235P 228P 410P 500P 546P 748P 907P 937P 1100P 910A 227P 813P 809A

B S L # S D B BL BL # SD LD D D B B B BS BL LD SD SD # D DS D DS # B B # # S D B B B SL L S D # BL # B B S L

D D # B # D B

0 1 0 1 0 0 1 2 1 1 1 1 1 1 1 1 1 0 3 3 1 1 1 0 0 0 1 1 0 0 2 0 0 0 1 0 0 1 0 0 0 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

D10 100 S80 733 D9S S80 757 757 M80 M80 757 D10 D10 757 M80 757 72S D10 733 734 L10 757 757 757 757 767 757 757 M80 L10 M80 M80 L10 72S 72S M80 73S 73S 72S M80 734 72S 757 M80 M80 D9S 757 757 757 757 D9S 757 757 D9S 72S 757 S80 757 S80 M80

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (13 of 67) [19/1/2003 1:40:22]

F F F F F F F F A F F F F F@ F F F F F F F F F F F F F@ F F F F F F F F FN F F F F F F F F F FN F F F F F F F F F FN F F F FN

Y Y Y Y Y Y Y Y Y Y Y Y C Y@ Y Y Y Y Y Y C Y Y Y Y C Y@ Y Y C Y Y C Y Y YN Y Y Y Y Y Y Y Y Y YN Y Y Y Y Y Y Y Y Y YN Y Y Y YN

B B B B B B B B Q B B B Y B@ B B B B B B Y B B B B Y B@ B B Y B B Y B B BN B B B B B B B B B BN B B B B B B B B B BN B B B BN

H M H Q H M H Q H Q H M M H M Q H K Q M M H M Q B M M@ Q@ Q M M Q M H M Q M H Q B Q M Q M Q M Q M H B Q M@ Q@ M H Q M B Q Q M Q M B Q Q M M Q M H H Q H Q M Q M Q H Q M H M Q Q M M Q M H M Q M Q M Q M M H M M H M Q MN QN H M H M H M MN QN

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

US1553 NW1850 US 758 NW1852 US 658 NW1854 US 698 NW1856 US 674 NW1858 US 726 NW1860 NW1862 US1682 US1615 NW1864 US1222 NW1866 NW1828 US1608 US1118 NW1870 DL 378 US1226 NW1872 NW1874 US1971 DL1862 NW1838 US 442 NW 563 AA1115 CO1515 AA 705 US 110 AA 417 AA 537 DL 527 US1665 AA 209 AA 473 DL 139 AA1679 DL 277 AA 885 DL 209 AA 609 DL 415 AA 177 DL 571 NW 285 NW 231 NW 11 NW 233 NW 337 US 95 NW 347 NW 239 NW 339 AA1679 TW 744

DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA

700A 700A 800A 800A 900A 900A 1000A 1000A 1100A 1100A 1200N 1200N 100P 100P 200P 200P 300P 300P 400P 400P 500P 500P 535P 600P 600P 700P 700P 755P 800P 800P 910A 1000A 1044A 1245P 630P 620A 740A 850A 1050A 1118A 1236P 1240P 159P 205P 259P 335P 459P 545P 559P 730P 650A 800A 945A 1100A 100P 255P 420P 545P 715P 159P 221P

BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS DEN DEN DEN DEN DEN DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DTW DTW DTW DTW DTW DTW DTW DTW DTW HOU JFK

826A 820A 929A 920A 1027A 1026A 1123A 1121A 1225P 1221P 125P 125P 223P 229P 325P 321P 427P 426P 526P 526P 627P 624P 655P 726P 725P 824P 827P 915P 924P 927P 1245P 111P 207P 430P 1000P 830A 955A 1052A 300P 144P 301P 553P 428P 415P 534P 551P 718P 753P 824P 932P 826A 938A 1122A 1239P 235P 555P 605P 728P 858P 631P 331P

B B B B B B S S S L L L L L S S S S S S D D S D D D D S S S BS # L LS S B B B L L L L S S G S D D D D B B # #

# S

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 0 0 0 1 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0

100 D9S D9S D9S 734 757 D9S D9S D9S M80 733 757 D9S 100 D9S 72S 734 D9S D9S 100 733 M80 72S 100 320 M80 734 72S D9S 733 M80 S80 733 S80 72S 757 757 757 M80 S80 757 757 S80 M80 S80 M80 S80 M80 757 757 320 320 757 D9S 320 D9S 757 320 320 S80 M80

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (14 of 67) [19/1/2003 1:40:22]

FN FN F F F F F F F F F F F F F F F F F F F F F Y F F FN F F FN F F A F F F F F F F F F F F F F F F F F FN FN F F F Y F F F F F

YN YN Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y B Y Y YN Y Y YN Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y YN YN C Y Y B Y Y Y Y C

BN HN QN BN M H B H Q B M H B H Q B M H B H Q B M H B H Q B M H B H Q B M H B M H B H Q B H M B M H B H Q B M H B M H B H Q B H Q B M H B M Q H M K B M H B M H BN HN QN B M Q B M H BN HN QN B M H B M Q Q H K B M Q B M B H M B H M B M Q B H Q B H M B H M B M H B H M B M Q B H M B M Q B H M B M Q B H M B M Q BN M H BN M H Y B M B M H B M H H Q M B M H B M H B M H B M H Y B Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

DL1864 TW 700 TW 10 CO 131 US 485 NW 551 US 890 DL 139 NW 337 AA1661 US 187 US1157 AA 615 NW 339 NW 311 US 809 AA 925 PA 977 NW1825 PA 495 US 768 AA 741 AA 433 PA 423 AA 897 PA2441 NW 111 NW 563 NW 319 NW 321 NW 711 UA 625 NW 323 NW 311 DL 139 DL 991 AA 273 UA 601 UA 603 UA 605 AA1115 UA 607 UA 609 UA 611 AA1701 AA 705 UA 615 UA 617 UA 619 AA 319 UA 621 UA 623 UA 625 AA 615 UA 627 AA 823 UA 629 AA1703 US 93 NW1269

DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA

305P 459P 833P 645A 700A 705A 1115A 1240P 100P 129P 215P 435P 629P 715P 815P 715A 723A 745A 859A 1045A 1220P 1259P 329P 330P 554P 630P 700A 910A 1200N 335P 515P 600P 630P 815P 1240P 755P 643A 700A 800A 900A 1000A 1000A 1100A 1200N 1200N 1245P 100P 200P 300P 359P 400P 500P 600P 629P 700P 738P 800P 900P 700A 745A

JFK JFK JFK LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MSP MSP MSP MSP MSP MSP MSP MSP MSY MSY ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD PHX PHX

440P 617P 938P 144P 1101A 1135A 316P 920P 527P 541P 608P 848P 1053P 1148P 1228A 1201P 959A 1020A 1134A 120P 425P 342P 607P 605P 831P 905P 845A 1054A 144P 522P 655P 908P 815P 950P 345P 1050P 744A 807A 900A 1002A 1102A 1059A 1159A 1259P 1257P 152P 201P 259P 358P 510P 501P 608P 708P 740P 815P 848P 858P 1010P 1120A 1210P

BL B B SL LD # # D # DS S S SB B # B # L L S S D D B B L S D D D # L # B B B B # # S L L L L S S S S D D D D # # B B

0 0 0 2 1 1 1 4 1 1 1 1 1 1 1 2 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1

72S 72S 72S M80 733 757 733 757 320 757 734 733 S80 320 320 734 72S 72S D9S 72S 73S 72S 727 72S 72S 72S 320 M80 320 M80 757 72S D9S 320 757 72S 757 757 72S 72S S80 72S 72S 72S S80 S80 72S 72S 72S S80 72S 757 72S S80 72S S80 72S S80 733 72S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (15 of 67) [19/1/2003 1:40:22]

F Y B M Q F Y B Q M F Y B Q M A Y Q H K F Y B H Q F Y B M H F Y B M F Y B M H F Y B M H F Y B H Q F Y B H Q F Y B H Q F Y B H Q F Y B M H FN YN BN M H F Y B H Q F Y B M H F Y B F Y B M H F Y B M Q F Y B H Q F Y B M H F Y B M H F Y B M Q F Y B M H F Y B M Q F Y B M H F Y B M H F Y B M H F Y B M H F Y B M H F Y B M Q F Y B M H FN YN BN M H F Y B M Q F@ Y@ B@ M@ Q@ F Y B M H F Y B M Q F Y B M Q F Y B M Q F Y B M H F Y B M Q F Y B M Q F Y B M Q F Y B M H F Y B M H F Y B M Q F Y B M Q F Y B M Q F Y B M H F Y B M Q F Y B M Q F Y B M Q F Y B M H F Y B M Q F Y B M H F Y B M Q FN Y B M H F Y B H Q F Y B M H

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

HP 91 US 350 US1211 HP 864 US1691 US1519 US 452 US1004 US1289 AA1083 AA 721 US1755 AA 363 US1726 DL 523 CO 891 AA 273 US 890 CO 215 US 187 NW 347 US1157 US 67 CO 629 TW 185 TW 447 TW 123 US 127 TW 99 TW 391 TW 55 TW 199 TW 537 TW 703 US 291 NW1841 DL 539 US 768 NW1845 US1158 UA 691 YV 236 CO1655 YV 12 UA 672 CO1707 UA 509 YV 688 CO 195 UA 466 YV 226 UA 245 CO1267 YV 608 UA 867 DL 402 NW 562 CO1532 UA 408 DL 317 DL1136

DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN

800A 340P 350P 440P 835A 1055A 400P 900P 650A 759A 1159A 100P 643P 655P 1045A 230P 643A 1115A 1200N 215P 420P 435P 520P 540P 700A 812A 943A 1150A 1210P 107P 359P 422P 529P 659P 830A 905A 905A 1220P 210P 620P 835A 840A 845A 1000A 1040A 1157A 1212P 1240P 251P 300P 530P 535P 604P 715P 836P 745A 930A 1020A 1024A 1100A 245P

PHX PHX PHX PHX PVD PVD PVD PVD RDU RDU RDU RDU RDU RDU SEA SEA SFO SFO SFO SFO SFO SFO SFO SFO STL STL STL STL STL STL STL STL STL STL TPA TPA TPA TPA TPA TPA ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ATL ATL ATL ATL ATL ATL

1215P 805P 840P 859P 950A 1206P 510P 1013P 758A 917A 109P 212P 800P 800P 230P 820P 1140A 513P 524P 816P 905P 1113P 938P 1012P 831A 945A 1112A 226P 137P 237P 527P 550P 702P 827P 1045A 1126A 120P 242P 432P 837P 949A 1030A 1002A 1155A 1150A 123P 134P 230P 410P 414P 720P 644P 733P 910P 947P 1226P 430P 317P 319P 400P 731P

S SD # S S

S V # # L SD BR SL # DS D # SD D B B S # L # S D D D B B S L S D S S

# #

# #

B SL L L L D

1 1 1 1 0 0 0 0 0 0 0 0 0 0 1 1 1 2 1 2 1 2 1 1 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 1 0 0 0 1 0 0 1 0 0 1 0 0 1 0 0 0 0

733 734 733 733 100 73S M80 73S 100 S80 S80 D9S 72S 734 72S M80 757 733 M80 734 757 733 733 733 M80 M80 M80 100 M80 M80 72S 72S D9S M80 M80 D9S 757 73S D9S 733 72S BE1 733 BE1 72S 733 733 BE1 73S 733 BE1 72S 73S BE1 733 757 M80 733 733 757 72S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (16 of 67) [19/1/2003 1:40:22]

F Y B F Y B F Y B Y B H F Y B F Y B F Y B F Y B F Y B F Y B F Y B Y B H F Y B F Y B F Y B A Y Q F Y B F Y B A Y H F Y B F Y B F Y B F Y B Y Q H F Y B F Y B F Y B Y B H F C Y F Y B F Y B F Y B F Y B F Y B F Y B FN YN BN F Y B F Y B F Y B F Y B F Y B Y Q L A Y Q Y Q L F Y B F A Y F Y B Y Q L F A Y F Y B Y F Y B A Y Q Y Q L F Y B F Y B F Y B A Y Q F Y B F Y B F Y B

H M H K H H H H H H H Q H H M H H M K H M H H K Q Q Q Q B Q Q Q Q Q H M M H M H M H H H M Q M H Q M

K

M H H M M M H M M M

Q K M Q Q H K Q Q Q

Q Q Q Q Q Q Q M M M M Q Q K M B Q H Q Q B M M M M Q M M M M M Q H Q Q H Q Q M K M Q H Q M H Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

CO 742 DL1644 UA 740 AA1260 NW 802 AA 796 DL1278 AA1012 US 884 UA 352 CO1234 CO 786 UA 354 CO 578 CO1182 DL 402 NW 560 CO1156 US1520 TW 240 AA 820 DL1278 UA 216 AA1166 AA1498 DL 544 AA1274 CO 840 UA 794 AA 730 DL 605 AA 50 CO 280 UA 450 CO 496 AA1326 DL 742 AA1195 UA 664 CO1760 DL 661 AA1698 NW1224 YX 522 UA 680 CO1704 NW1226 NW 564 NW1228 UA 458 UA 496 CO1650 UA 514 UA 908 TW 870 UA 263 UA 193 CO1245 CO 479 UA 169

DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN

414P 615P 633P 935A 1000A 307P 615A 619A 805A 1029A 1035A 104P 335P 401P 729P 745A 755A 855A 1055A 419P 435A 615A 645A 704A 845A 940A 1023A 1025A 1031A 1150A 120P 120P 120P 309P 405P 405P 505P 605P 633P 733P 750P 845P 810A 940A 1029A 1034A 1115A 130P 250P 307P 645P 728P 1044A 639P 948A 645A 845A 849A 1200N 1215P

ATL ATL ATL BNA BNA BNA BOS BOS BOS BOS BOS BOS BOS BOS BOS DCA DCA DCA DCA DCA DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW HOU HOU JFK LAX LAX LAX LAX LAX

909P 1051P 1123P 109P 242P 647P 210P 133P 322P 420P 444P 701P 918P 1057P 123A 259P 249P 341P 544P 1108P 728A 900A 933A 1001A 1152A 1224P 127P 120P 115P 256P 415P 424P 407P 606P 659P 716P 750P 858P 924P 1025P 1034P 1138P 1251P 340P 310P 327P 404P 744P 739P 744P 1128P 1216A 207P 1002P 325P 809A 1011A 1021A 129P 136P

D D D R S D B BL BS L L L D D D BL SL B L # B B B B S L # L L L # S # # D D D D S S B BS L L L S D D D D L D L B B B L L

0 0 0 0 1 0 1 1 1 0 0 0 0 1 0 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0

733 72S 737 S80 D9S S80 M80 S80 72S 757 M80 M80 757 733 733 757 M80 M80 72S M80 S80 M80 735 S80 S80 72S S80 73S 72S S80 M80 S80 M80 737 733 S80 72S S80 72S D9S 72S S80 D9S DC9 733 72S 72S M80 72S 72S 733 72S 737 737 M80 735 D10 M80 AB3 D10

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (17 of 67) [19/1/2003 1:40:22]

F F F F F F F@ F F F A F F Y A F F A F F FN FN F F F F F F F F Y F F F A F F F F A FN F F Y F F F F F F F F F F F F F A F F

A Y Q H Y B M Q Y B M Q Y B H M Y B M H Y B H M Y@ B@ M@ Q@ Y H B M Y B H Q Y B M Q Y Q H K A Y Q H Y B M Q H K B V Y Q H K Y B M Q Y B M H Y Q H K Y B H Q Y B Q M YN B Q M YN BN MN QN Y B M Q Y B Q M Y B Q M Y B M Q Y B Q M A Y Q H Y B M Q Y B Q M B M Q H Y B Q M A Y Q H Y B M Q Y Q H K Y B Q M Y B M Q Y B Q M Y B M Q Y Q H K YN BN MN QN YN B Q M Y B M H H M B K Y B M Q A Y Q H Y B M H Y B M H Y B M H Y B M Q Y B M Q A Y Q H Y B M Q Y B M Q Y B Q M Y B M Q Y B M Q Y Q H K A Y Q H Y B M Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

CO 347 UA 497 DL 309 UA 857 CO1629 HP 38 UA 539 UA 239 CO 211 UA 230 UA 382 CO1212 UA 986 NW 560 NW 562 UA 982 CO1722 NW 568 NW 564 UA 636 CO1674 NW 566 UA 572 CO1754 UA 956 CO1568 CO1212 CO 530 UA 434 AA1012 UA 220 ZK 76 AA1276 UA 226 UA 230 CO1514 UA 228 ZK 403 AA 768 UA 940 UA 730 CO 806 UA 234 AA1362 UA 246 CO 136 NW 566 AA 388 UA 238 CO 598 UA 242 UA 270 CO1602 AA 768 UA 494 CO 780 HP 844 UA 847 CO 735 HP 854 UA 419

DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN

244P 252P 425P 545P 600P 603P 615P 826P 826P 921A 310P 403P 700A 755A 930A 1033A 1034A 1100A 130P 307P 402P 505P 643P 719P 1033A 1035A 403P 409P 635P 619A 705A 845A 909A 920A 921A 1034A 1050A 1120A 1141A 1153A 1231P 115P 315P 330P 345P 412P 505P 545P 655P 732P 850P 1038A 1040A 1141A 330P 401P 645A 835A 855A 1140A 1155A

LAX LAX LAX LAX LAX LAX LAX LAX LAX MIA MIA MIA MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSY MSY MSY MSY MSY ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD PHL PHL PHL PHL PHL PHX PHX PHX PHX PHX

410P 414P 710P 700P 724P 830P 737P 945P 947P 604P 1011P 1102P 945A 1045A 1220P 124P 135P 145P 418P 602P 703P 755P 938P 1014P 212P 216P 739P 1151P 1016P 945A 1017A 315P 1245P 1238P 1245P 152P 210P 520P 309P 309P 350P 432P 626P 717P 703P 739P 947P 915P 1004P 1046P 1157P 356P 416P 726P 850P 940P 842A 1036A 1100A 141P 144P

S S SD D D S D # LS D D B S S L L S S S S S D # L L D D B B R L L L L L L L L D D D D S D D S # L L LD D D B # S L

0 0 1 0 0 1 0 0 0 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 5 0 0 0 0 0 4 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0

M80 733 757 D10 AB3 733 737 757 733 733 72S 733 737 M80 M80 72S M80 D9S M80 737 72S 72S 733 73S 735 733 733 72S 737 S80 D10 BE1 S80 72S 733 M80 D10 BE1 757 D10 757 M80 D10 S80 D10 72S 72S S80 D10 M80 72S D10 M80 757 757 M80 733 733 M80 73S 72S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (18 of 67) [19/1/2003 1:40:22]

A Y Q F Y B F Y B F Y B A Y Q Y B H F Y B F Y B F A Y F F Y B Y H K F Y B FN YN BN F Y B F Y B F A Y F Y B F Y B F Y B F A Y F Y B F Y B A Y Q F Y B A Y Q Y H K A Y Q F Y B F Y B F Y B Y M H F Y B F Y B F Y B F A Y F Y B Y M H F Y B F C Y F Y B F A Y F Y B F Y B F Y B F A Y F Y B F Y B F Y B F A Y F Y B F Y B A Y Q F Y H F Y B F A Y Y B H F Y B A Y Q Y B H F Y B

H M M M H K M M Q

K Q Q Q K Q H Q H

M B M M M M Q M M M Q M M H M H B H M H M Q H M M Q M Q H B M Q M H M Q M H M Q M M H B M Q K M H K M

Q V Q H H Q H H H Q H H Q K H K V K H M Q V M Q Q H Q V M M Q H Q M Q H H M Q H Q Q K M Q H Q Q K Q Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

CO1247 CO 201 HP1523 HP 97 UA 439 CO 249 HP 840 UA 695 CO 467 AA1362 DL1136 AA1326 HP 844 UA 293 CO1257 UA 319 UA 455 CO1239 CO1259 UA 427 UA1547 CO 171 HP 840 CO 255 UA 227 UA 681 UA 819 CO 395 UA 315 UA 343 CO 445 HP 603 UA 247 UA 347 UA 207 UA1727 CO 173 CO1765 UA 861 TW 402 TW 450 UA 854 TW 118 UA 470 TW 240 UA 476 CO1568 DL 317 US 529 DL 581 AA1067 YV 81 AA1257 DL 527 AA 329 DL 545 AA 305 AA 163 DL 441 AA1111

DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW

1207P 245P 320P 450P 519P 605P 655P 824P 825P 330P 245P 405P 645A 835A 858A 930A 1157A 1209P 247P 325P 543P 630P 655P 828P 842P 640A 837A 850A 930A 1149A 1205P 105P 150P 430P 519P 530P 635P 831P 851P 700A 855A 1046A 100P 305P 419P 637P 1035A 1100A 115P 812A 921A 1030A 1118A 1147A 227P 321P 358P 633P 642P 1000P

PHX PHX PHX PHX PHX PHX PHX PHX PHX PVD RDU RDU SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO STL STL STL STL STL STL STL TPA TPA TPA ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ

200P 436P 519P 649P 707P 810P 854P 1008P 1020P 1058P 1010P 1139P 1157A 1027A 1102A 1128A 144P 205P 443P 520P 728P 829P 1158P 1014P 1032P 820A 1016A 1028A 1116A 130P 157P 440P 316P 609P 830P 703P 813P 1010P 1022P 1000A 1203P 146P 404P 601P 727P 933P 519P 620P 838P 905A 1003A 120P 1207P 1230P 315P 405P 446P 724P 740P 1043P

# S # D # S

# D D S B B B L L S D D D S # B B B B L L S S D D D # # B B L # S D D L LS LD # V L L

D D

0 0 0 0 0 0 0 0 0 1 1 1 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 1 1 0 0 2 0 0 0 0 0 0 0 0

72S 733 733 733 72S 72S 73S 72S M80 S80 72S S80 733 D10 72S 735 D10 M80 72S 733 D10 AB3 73S M80 757 733 D10 72S 737 757 AB3 733 72S 735 735 757 AB3 733 757 72S 72S 733 72S 72S M80 72S 733 757 72S M80 72S BE1 S80 757 S80 72S 72S S80 M80 72S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (19 of 67) [19/1/2003 1:40:22]

F A A Y F Y Y B F Y F A Y B F Y F A F Y F Y F Y Y B F Y F A F Y F Y F A A Y F Y F Y F A Y B F A F Y F Y F Y F A Y B F Y F A Y B F Y F Y F Y F Y A Y F A F Y F Y F Y F Y F Y F Y F Y F Y A Y F Y F Y F Y F Y Y Q F Y F Y F Y F Y F Y F Y F Y FN YN

Y Q B H B Y H B Y H B H H B Y B B Y Q B B Y H Y B B B Y M B Y H B B B B Q Y B B B B B B B B Q B B B B L B B B B B B B B

Q H H K M Q K M Q B M B K M Q M M Q H M M Q K Q M M M Q Q M Q K M M M M H Q M Q Q M Q M Q M H

H K K Q Q H Q Q H M Q M Q Q H Q Q H K Q Q H Q H Q Q Q H H Q H Q Q H Q H K H Q M M Q M Q M Q K

H M M H M M M M M M M M

Q Q H M H Q H Q H H Q H

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

DL 333 DL 16 DL 500 DL 636 DL1198 DL 542 AA 864 DL1190 DL 408 DL 353 AA 752 DL 790 DL 784 DL 508 DL1007 DL1046 DL 472 DL1028 DL 904 DL 833 DL 941 DL1094 AA 504 DL1066 DL 756 DL 335 DL1074 DL 618 DL1264 DL 783 AA 482 DL 688 DL1544 DL 394 DL 940 DL 722 DL 556 AA 222 DL1061 DL 592 DL 260 DL 908 DL 832 AA 496 DL 154 DL1008 DL1080 AA1144 AA1280 AA 410 DL 784 AA1444 AA1086 DL1192 AA1406 DL1078 AA 352 AA 440 DL 478 DL 996 AA 728

DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW

1028P 615A 615A 615A 640A 641A 656A 805A 806A 809A 816A 940A 1020A 1041A 1125A 1125A 1138A 1140A 1143A 1144A 1145A 1201P 1245P 100P 100P 106P 108P 115P 122P 141P 238P 300P 304P 305P 314P 450P 453P 516P 557P 557P 558P 641P 645P 804P 808P 1027P 1138P 645A 807A 814A 1020A 1102A 1251P 110P 239P 453P 520P 803P 831P 610A 635A

ABQ ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BOS BOS

1110P 918A 1055A 1100A 1233P 1051A 953A 1105A 1235P 1237P 1124A 1242P 228P 227P 410P 409P 235P 401P 355P 406P 405P 359P 340P 410P 559P 556P 556P 554P 549P 548P 538P 600P 730P 755P 738P 744P 935P 817P 1033P 1039P 1054P 1101P 942P 1102P 1105P 115A 223A 826A 958A 1245P 1200N 1246P 240P 302P 423P 635P 722P 956P 1015P 1146A 1058A

B # # # B B B S # B S S L L L L L L S L L S S S L S S S D D D D S D S S D # #

B B

S L L D D

B B

0 0 2 2 3 1 0 0 1 1 0 0 1 1 1 2 0 1 1 1 1 1 0 0 2 2 2 1 1 1 0 0 1 1 1 0 2 0 2 2 2 1 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 1 0

72S L15 M80 M80 M80 M80 S80 757 M80 M80 72S 763 M80 D9S M80 M80 757 M80 73S 73S M80 M80 S80 L10 72S 72S M80 757 M80 M80 S80 L10 D9S M80 73S L10 73S 727 73S 73S M80 D9S M80 727 767 D9S M80 S80 S80 727 M80 S80 S80 72S S80 73S 757 757 72S 72S S80

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (20 of 67) [19/1/2003 1:40:22]

FN F F F@ F@ F@ FN F Y F FN F F F F F F F F F F F F F F F F F F F F F F F F F F@ F F@ F@ F@ F@ F FN FN FN FN F F F F F F F F F F F F F F

YN Y Y Y@ Y@ Y@ YN Y B Y YN Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y@ Y Y@ Y@ Y@ Y@ Y YN YN YN YN Y Y Y Y Y Y Y Y Y Y Y Y Y Y

BN B B B@ B@ B@ B B M B B B B B B B B B B B B B B B B B B B B B B B B B B B B@ B B@ B@ B@ B@ B B BN BN BN B B M B B B B B B B B B B B

MN M M M@ M@ M@ H M Q M H M M M M M M M M M M M H M M M M M M M H M M M M M M@ H M@ M@ M@ M@ M H MN MN MN M M B M M M M M M M M M M H

QN Q H Q@ Q@ Q@ M Q H Q M Q Q Q Q Q Q Q Q Q Q Q M Q Q Q Q Q Q Q M Q Q Q Q Q Q@ M Q@ Q@ Q@ Q@ Q M QN QN QN H H H Q H H Q H Q H H Q Q M

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AA1364 DL1278 AA 108 DL 358 US1473 AA 634 DL1620 AA 154 AA 568 DL 874 AA 804 AA1284 US 349 DL 400 AA 236 DL1040 DL 522 AA1256 AA1322 DL 394 DL 660 AA 846 AA 834 DL 462 UA1483 AA1697 CO 277 DL 888 AA 861 AA1281 CO1219 UA 985 AA 419 DL 233 AA 527 DL 309 UA 851 AA 51 CO 885 AA 193 CO 233 UA 577 DL 509 AA1631 AA 313 DL1055 AA 198 NW 690 NW1560 AA 382 DL 672 NW 692 DL 262 UA 446 AA1132 NW 696 DL 588 NW 694 AA1168 AA 212

DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW

813A 941A 103P 107P 140P 223P 507P 519P 656A 820A 827A 924A 925A 955A 1051A 100P 102P 106P 243P 305P 500P 516P 800P 824P 700A 700A 700A 807A 811A 934A 1010A 1014A 1108A 1140A 222P 301P 340P 355P 432P 635P 645P 646P 656P 803P 953P 1033P 700A 800A 915A 940A 949A 1220P 116P 206P 232P 400P 501P 505P 515P 521P

BOS BOS BOS BOS BOS BOS BOS BOS DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW

1253P 210P 525P 555P 757P 644P 944P 948P 1033A 1205P 1214P 105P 255P 140P 231P 705P 459P 447P 628P 1025P 845P 904P 1136P 1205A 802A 810A 805A 900A 913A 1036A 1104A 1113A 1207P 1237P 320P 355P 437P 455P 525P 740P 734P 748P 755P 904P 1049P 1130P 1019A 1118A 1236P 107P 120P 350P 455P 708P 559P 728P 830P 847P 849P 1040P

B B # L SD G D D B B B R B B L S L # S D D D S S B B S B B S S S L

# # D # D D

B B B R B L L # S D D D D D

0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 2 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1

S80 M80 D10 757 733 757 M80 D10 S80 72S 757 S80 M80 M80 S80 72S M80 757 S80 M80 757 757 757 M80 733 S80 733 72S S80 S80 D9S 735 S80 M80 S80 757 733 S80 73S S80 733 737 72S S80 S80 M80 72S D9S D9S S80 73S D9S M80 72S S80 D9S 72S D9S S80 D10

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (21 of 67) [19/1/2003 1:40:22]

F F F F F F F F F F F F F F F F F F F F F F F F F F A FN F F A F F F F F F F A F A F FN F FN FN F FN FN F F F F F F F F FN F F

Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y YN Y Y Y Y Y Y Y Y Y Y Y Y Y Y YN YN YN YN Y YN YN Y Y Y Y Y Y Y Y YN Y Y

B H M B M Q B H M B M Q B H Q B H M B M Q B H M B H M B M Q B H M B H M B H Q B M Q B H M B M Q B M Q B H M B H M B M Q B M Q B H M B H M B M Q B M Q B Q M Q H K BN MN QN B Q M B Q M Q H K B M Q B Q M B M Q B Q M B M Q B M Q B Q M Q H K B Q M Q H K B M Q BN MN QN B Q M B Q M BN MN QN B H M BN M H BN M H B H M B M Q B M H B M Q B M Q B H M B M H B M Q BN M H B H M B M H

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AA 158 DL1177 AA 8 DL 408 AA 371 AA 803 TW 130 DL 877 AA1344 AA 856 DL 946 AA1179 AA1679 AA1509 DL 853 AA 401 DL1228 AA 926 TW 130 DL1870 AA 862 AA1165 AA 509 DL 131 DL 859 DL1715 AA 227 AA 333 DL 285 AA 625 AA 281 DL 819 DL 309 DL 179 AA 49 AA 603 AA 389 DL1215 DL 139 DL 161 AA 717 AA1261 DL 279 DL1700 AA 857 AA 614 DL 546 AA1614 DL 472 AA 880 DL 174 AA 498 PA 737 AA 734 AA1328 DL 558 NW 516 AA 832 NW 400 AA 629 NW 402

DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW

757P 617A 652A 806A 810A 1050A 1100A 1152A 1251P 230P 301P 354P 514P 630P 651P 951P 1027P 1116P 1100A 1155A 1249P 700A 805A 806A 812A 812A 924A 1120A 1151A 1253P 239P 300P 301P 340P 400P 514P 642P 650P 652P 710P 800P 959P 1050P 640A 645A 815A 940A 1100A 1138A 1257P 107P 225P 515P 519P 805P 846P 630A 809A 830A 1055A 1115A

DTW HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU JFK JFK JFK LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MSP MSP MSP MSP MSP

1111P 715A 756A 910A 918A 1157A 1200N 1255P 159P 339P 403P 458P 631P 738P 755P 1057P 1125P 1215A 450P 440P 525P 811A 921A 920A 1019A 1016A 1017A 1219P 100P 200P 337P 508P 710P 440P 504P 625P 746P 909P 920P 820P 907P 1056P 1158P 1010A 1007A 1151A 110P 235P 515P 439P 445P 604P 901P 902P 1136P 1215A 836A 1029A 1038A 109P 130P

# S # S #

L L L R R B BS B R L L L S S SD S D D D D D D S S S B B B B L L L L S D D # # B B B L L

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 1 0 0 0 0 0 1 2 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0

72S D9S S80 M80 72S S80 72S D9S S80 S80 D9S 72S S80 S80 73S S80 73S 72S 72S 72S 72S S80 S80 L10 72S 72S D10 D10 763 S80 D10 72S 757 L10 D10 S80 D10 72S 757 L10 S80 D10 763 72S 757 D10 72S 72S 757 72S 72S 72S 72S 757 757 L10 D9S S80 D9S S80 D9S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (22 of 67) [19/1/2003 1:40:22]

F FN F F F F F F F F F F F F F FN FN FN F F F F F F F F F F F F F F F F F F F F F F F FN FN F F F F F F F F F F F F F F F F F FN

Y YN Y Y Y Y Y Y Y Y Y Y Y Y Y YN YN YN Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y YN YN Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y YN

B H M BN MN QN B M H B M Q B M H B M H B Q M B M Q B M H B M H B M Q B M H B M H B M H B M Q B M H BN MN QN B M H B Q M B M Q B H M B H M B H M B M Q B M Q B M Q B H M B H M B M Q B H M B H M B M Q B M Q B M Q B H M B H M B H M B B M Q B M Q B H M B H M BN MN QN B M Q B H M B H M B M Q B H M B M H B H M B B B B B B B B B BN

H M H H M M M M M M

M Q M M Q H H H H H

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AA 438 AA 760 NW 404 NW 406 AA 822 DL 824 AA 548 AA1342 DL1042 AA 616 DL 618 DL1716 AA 374 AA 654 DL1720 AA1542 DL 694 AA 580 UA 224 AA 470 DL 310 AA 360 AA 784 UA 252 DL 124 DL 335 AA1554 UA 446 AA1502 DL 721 AA 224 DL1240 UA 280 AA 212 AA 476 UA 776 AA 150 AA1044 AA 456 AA 124 AA 360 DL1006 YX 305 US 96 AA1010 DL1295 AA1388 AA 586 AA1074 DL 252 HP 622 AA1349 DL 995 AA 475 AA1447 DL 444 HP 834 DL1019 AA1441 HP 862

DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW

1251P 236P 425P 550P 756P 617A 635A 812A 942A 102P 115P 448P 514P 806P 824P 1111P 1149P 700A 805A 811A 820A 925A 1054A 1057A 100P 106P 108P 206P 233P 311P 352P 450P 508P 521P 645P 700P 808P 1000P 1110P 657A 925A 1003A 1050A 1220P 1257P 112P 232P 513P 755P 845P 705A 805A 814A 930A 1112A 1153A 240P 317P 356P 514P

MSP MSP MSP MSP MSP MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX

305P 459P 637P 759P 1009P 735A 751A 945A 1105A 232P 250P 605P 643P 933P 950P 1233A 105A 915A 1013A 1019A 1040A 1131A 1256P 101P 314P 750P 315P 406P 434P 520P 606P 705P 725P 741P 915P 912P 1015P 1204A 114A 1049A 303P 210P 430P 546P 459P 530P 628P 913P 1141P 1245A 836A 935A 945A 1100A 1244P 110P 412P 440P 517P 646P

L D D # # # #

S S S

B B B B V L L L SD # # # S S D D D D D #

B VL B L L L L S D # S S B B V L L S S S S

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

S80 S80 D9S D9S S80 M80 72S S80 72S S80 757 757 S80 S80 757 S80 757 757 72S D10 72S S80 757 72S 72S 72S D10 72S D10 72S 757 D9S 733 D10 72S 733 D10 72S S80 S80 S80 73S DC9 734 S80 M80 S80 S80 S80 757 73S S80 72S S80 S80 72S 733 72S D10 733

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (23 of 67) [19/1/2003 1:40:22]

F F F F F FN F F F F F F F F F F FN F F F F F F F F F F F F F F F F F F F F FN FN F F F Y F F F F F F FN Y F F F F F F F F F

Y B M H Y B M H Y B M H Y B M H Y B M H YN BN MN QN Y B M H Y B M H Y B M Q Y B M Y B M Q Y B M Q Y B M Y B M Y B M Q YN B M H YN BN MN QN Y B M H Y B M Q Y B M H Y B M Q Y B M H Y B M H Y B M Q Y B M Q Y B M Q Y B M H Y B M Q Y B M H Y B M Q Y B M H Y B M Q Y B M Q Y B M H Y B M H Y B M Q Y B M H YN B M H YN B M H Y B H M Y M H B Y B M Q H M B Y B H Q Y B H M Y B M Q Y B H M Y B H M Y B H M YN BN MN QN B H K Q Y B H M Y B M Q Y B H M Y B H M Y B M Q Y B H K Y B M Q Y B H M Y B H K

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AA 25 DL 219 HP 636 AA 503 DL1705 UA1098 US 794 AA1254 AA 320 DL1114 AA 840 AA 810 AA 844 AA 852 AA1326 DL 926 DL1717 HP 847 AA1339 AA 395 DL 703 DL1283 DL 887 DL1735 AA 251 HP 843 AA 217 DL1709 AA 583 AA 203 DL 825 UA 397 AA 125 AA1143 DL 669 DL 307 AA 71 AA1119 AA 79 DL 691 AA1191 DL 487 TW 563 TW 522 AA1640 AD1215 TW 278 TW 695 AA 344 AD1274 TW 674 AA 464 TW 487 AA 216 TW 664 AA 328 AA1474 DL 276 AA1378 DL 790 AA 206

DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW

637P 651P 734P 957P 1022P 145P 420P 814A 930A 946A 1057A 1251P 241P 517P 810P 823P 808A 845A 928A 1055A 1143A 1203P 314P 345P 403P 415P 634P 650P 953P 810A 811A 840A 934A 1055A 1148A 314P 402P 509P 638P 704P 958P 1024P 641A 801A 812A 935A 1025A 1235P 1252P 225P 230P 239P 354P 515P 537P 759P 1115P 624A 658A 940A 1106A

PHX PHX PHX PHX PHX PVD PVD RDU RDU RDU RDU RDU RDU RDU RDU RDU SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL TPA TPA TPA TPA

801P 820P 903P 1116P 1140P 722P 1025P 1149A 109P 126P 231P 418P 612P 904P 1139P 1155P 1120A 111P 1146A 104P 140P 335P 510P 710P 625P 852P 858P 900P 1208A 947A 955A 1018A 1125A 1225P 125P 450P 543P 705P 830P 855P 1143P 1159P 827A 945A 1006A 120P 1207P 215P 229P 610P 413P 420P 537P 704P 722P 934P 1251A 930A 1009A 305P 218P

D D S # LS S B R B L L D D # B S R L L L D SD D S D D B B B R L L S D D D D S S B B B

# L

S D

B B S L

0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 1 1 0 0 0 1 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 3 0 0 0 0 0 0 0 0 0 1 0

D10 72S 733 72S 72S 733 M80 72S S80 73S S80 72S S80 S80 S80 M80 72S 73S 757 72S 72S 757 72S 757 S80 73S S80 757 S80 757 767 72S S80 D10 763 767 757 S80 767 72S S80 757 D9S DC9 S80 SWM D9S D9S S80 SWM D9S S80 D9S S80 D9S S80 72S 72S 72S 763 72S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (24 of 67) [19/1/2003 1:40:22]

F F Y FN FN F F F F F F F F F F F F Y F F F F F F F Y F F FN F F F F F Y F F F F F FN FN F F F Y F F F Y F F F F F F FN F F F F

Y B H M Y B M Q B H K Q YN B H M YN BN MN QN Y B M Q Y B H Q Y B M H Y B M H Y B M Q Y B M H Y B M H Y B M H Y B M H Y B M H Y B M Q Y B M Q B H K Q Y B H M Y B H M Y B M Q Y B M Q Y B M Q Y B M Q Y B H M B H K Q Y B H M Y B M Q YN B H M Y B H M Y B M Q Y B M Q Y B H M Y B H M B M Q H Y B M Q Y B H M Y B H M Y B H M Y B M Q YN B H M YN BN MN QN Y B Q M Y B Q M Y B M H K Q V H Y B Q M Y B Q M Y B M H K Q V H Y B Q M Y B M H Y B Q M Y B M H Y B Q M Y B M H YN B M H Y B M Q Y B M H Y B M Q Y B M H

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

DL 584 AA1100 AA 178 DL 364 AA 618 DL1158 AA 305 DL 201 NW 491 DL 821 NW 495 NW 799 DL 501 DL 817 NW 499 DL 951 NW 795 DL 818 NW1565 AA1663 NW1567 NW1569 AA1057 WN 413 NW1523 NW1497 AA1543 US 852 NW 380 NW 932 NW 386 NW 330 NW 48 NW 390 NW 394 NW 338 NW 232 NW1109 NW 234 NW 236 NW 12 NW 286 NW 417 NW 250 UA 307 CO 395 NW1223 UA 701 NW1219 CO 249 UA 699 NW1225 AA1543 AA1433 AA1541 DL1271 NW 691 DL 940 NW 693 AA 305

DFW DFW DFW DFW DFW DFW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW

116P 235P 518P 524P 755P 835P 105P 550A 715A 845A 1030A 1225P 1240P 405P 440P 545P 645P 750P 720A 840A 935A 1210P 203P 300P 500P 640P 730P 700A 730A 1025A 140P 310P 455P 655P 845P 720A 1025A 1215P 135P 315P 440P 500P 655P 835P 657A 700A 915A 945A 1230P 418P 646P 650P 730P 655A 810A 900A 930A 1220P 1230P 105P

TPA TPA TPA TPA TPA TPA ABQ ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL BNA BNA BNA BNA BNA BNA BNA BNA BNA BOS BOS BOS BOS BOS BOS BOS BOS DCA DCA DCA DCA DCA DCA DCA DCA DCA DEN DEN DEN DEN DEN DEN DEN DEN DEN DFW DFW DFW DFW DFW DFW DFW

440P 551P 838P 840P 1101P 1155P 446P 737A 915A 1029A 1229P 223P 235P 554P 636P 733P 841P 937P 800A 912A 1019A 1249P 233P 440P 540P 720P 806P 1017A 918A 1215P 330P 503P 649P 850P 1037P 846A 1151A 140P 259P 440P 607P 626P 821P 959P 810A 810A 1031A 1058A 143P 516P 755P 810P 1034P 836A 954A 1033A 1137A 203P 238P 250P

L # D D # S # B B S L L

D D S # S

S BS B S

# S B #

# # B B B L L S D D S B B B B L L #

0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0

757 72S 72S 767 72S 763 72S 72S 72S 757 DC9 D9S M80 M80 D9S M80 DC9 757 D9S S80 D9S D9S S80 73S D9S DC9 S80 D9S 72S 320 757 320 D10 72S D10 757 320 D9S 757 D9S M80 320 757 320 733 72S 72S 733 72S 72S 733 D9S S80 S80 72S M80 D9S 73S D9S 72S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (25 of 67) [19/1/2003 1:40:22]

F F F F F F F FN FN F F FN F F F F F F FN F FN FN F Y F F F FN F F F F F F FN FN F F F F F F F F F F F F F A F F F F F F FN F FN F

Y B M Q Y B M H Y B M H Y B M Q Y B M H Y B M Q Y H B M YN BN MN QN YN BN M H Y B M Q Y B M H YN BN M H Y B M Q Y B M Q Y B M H Y B M Q Y YN Y YN YN Y K Y Y Y YN Y Y Y Y C Y YN YN Y Y Y Y Y Y Y Y Y A Y Y Y Y Y Y Y Y Y Y YN Y YN Y

B M Q BN M H B H M BN M H BN M H B H M L B Q B M H B M H B H M BN HN QN B M H B M H B M H B M H Y B M B M H BN M H BN M H B M H B M H B M H B M H B M H B M H B M H B M H B M Q Y Q H B M H B M Q B M Q H K B M Q H B B B BN B BN B

B M M M M M M M

M H H Q H Q H H

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

NW1185 AA1509 DL 645 DL 817 AA1699 NW 695 NW 697 WN 531 NW 271 NW 273 WN 413 AA1509 WN 505 NW 275 TW 766 NW 333 NW 335 NW 337 NW 51 AA 435 NW 339 PA 437 AA1551 NW 995 NW 993 NW 997 US 169 PA 421 US 537 NW 999 NW 123 NW 101 NW 115 NW 165 NW 743 NW 745 UA 567 NW 747 NW1073 NW 751 NW 753 NW 755 NW 757 NW1473 NW 297 NW 123 UA 223 AA 73 NW 115 UA 723 NW 165 AA 117 UA 461 NW 479 AA 847 UA 567 NW 205 NW 171 AA 835 UA 135 NW 173

DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW

310P 350P 400P 405P 645P 655P 940P 700A 930A 1225P 300P 350P 440P 650P 300P 925A 1225P 315P 700P 815P 940P 715A 755A 910A 1030A 1210P 225P 235P 525P 640P 700A 730A 800A 910A 925A 1235P 202P 310P 430P 540P 700P 850P 1000P 935A 650P 700A 710A 745A 800A 830A 910A 1035A 1110A 1215P 138P 202P 310P 435P 500P 520P 645P

DFW DFW DFW DFW DFW DFW DFW HOU HOU HOU HOU HOU HOU HOU JFK LAX LAX LAX LAX LAX LAX MIA MIA MIA MIA MIA MIA MIA MIA MIA MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSY MSY ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD

515P 540P 550P 940P 838P 903P 1142P 1040A 1146A 244P 655P 738P 910P 910P 449P 1130A 235P 527P 900P 1129P 1148P 1016A 1205P 1215P 135P 311P 645P 536P 1038P 942P 920A 827A 1027A 1127A 1021A 133P 417P 358P 635P 638P 804P 942P 1052P 1119A 947P 714A 722A 801A 816A 839A 920A 1051A 1120A 1235P 145P 215P 325P 450P 513P 550P 703P

S S D D

B L S D B L S D S S B B B L L S S S D S S S

S

S S S

B D S S # S S

0 0 0 3 0 0 0 2 0 0 2 1 3 0 0 0 0 0 0 1 0 0 1 0 0 0 1 0 1 0 1 0 1 1 0 0 1 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

D9S S80 73S M80 S80 D9S D9S 733 D9S D9S 73S S80 733 D9S 72S D10 757 320 D10 D10 320 72S 72S D9S D9S 72S 733 72S 73S D9S 320 D10 320 72S 757 D10 72S 72S DC9 744 D10 M80 D10 D9S 72S 320 D10 D10 320 72S 72S 767 D10 72S S80 72S 72S 72S 757 D10 72S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (26 of 67) [19/1/2003 1:40:22]

FN F F F@ F F F Y FN FN Y F Y F F F F F F F FN F F F F FN F F F B F F F F F F F F F F F FN FN F F F F F F F F F F F F F F F F F F

YN BN M H Y B M H Y B M Q Y@ B@ M@ Q@ Y B M H Y K YN YN K Y K Y Y Y Y Y C YN YN Y Y Y B YN Y Y Y

B L BN BN L H L B B B B B Y H BN B B B M BN B B B

M B M M B B B M Q M M M B B M M M M H M H M H

H Q H H Q M Q H M H H H M M H Q H H

Y B Y B Y B Y B Y B Y B Y B Y B Y Y B Y B YN BN YN BN Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B

M M M M M M M M

H H H H H H H H

M M M M M M M M M M M M M M M M M M M M M M

H H H H H H H Q V H Q H V Q H V H H H V Q H

H Q Q Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AA 435 UA 871 NW 175 NW1151 US 100 NW1194 NW 200 US1664 NW 202 NW 204 NW 340 US 537 NW 210 NW 101 TW 169 NW 247 NW 249 CO 249 DL 647 NW 391 NW 253 NW1440 NW1444 NW1476 NW 498 US 100 AA1551 AA1353 AA1529 US 376 NW 575 UA 461 NW 839 NW 583 NW 581 CO 395 AA1663 NW 929 AA 117 NW 343 NW 345 UA 135 NW 347 NW 349 WN 531 TW 169 NW 461 TW 337 NW 463 TW 91 NW 367 TW 395 NW1559 WN1735 TW 345 NW 484 NW 476 NW 480 CO 441 NW 482

DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW

815P 830P 835P 940P 700A 715A 1025A 1120A 135P 320P 440P 525P 840P 730A 735A 920A 1230P 418P 450P 700P 945P 715A 130P 435P 825P 700A 755A 1150A 640P 700P 925A 1110A 1210P 320P 700P 700A 840A 930A 1035A 1220P 320P 520P 655P 945P 700A 735A 920A 1042A 1215P 145P 320P 456P 640P 740P 807P 800A 920A 1230P 535P 655P

ORD ORD ORD ORD PHL PHL PHL PHL PHL PHL PHL PHL PHL PHX PHX PHX PHX PHX PHX PHX PHX PVD PVD PVD PVD RDU RDU RDU RDU RDU SEA SEA SEA SEA SEA SFO SFO SFO SFO SFO SFO SFO SFO SFO STL STL STL STL STL STL STL STL STL STL STL TPA TPA TPA TPA TPA

828P 846P 854P 952P 828A 843A 1152A 1235P 301P 447P 615P 653P 1008P 1156A 1123A 1146A 254P 810P 835P 923P 1202A 854A 309P 616P 959P 1028A 925A 119P 816P 1025P 1126A 233P 210P 514P 856P 1028A 1250P 1142A 227P 233P 533P 928P 905P 1151P 830A 812A 1005A 1127A 1257P 228P 404P 542P 728P 910P 845P 1037A 1200N 312P 1055P 934P

B S S S

S SB B B L # D D S S

BS B L D D B L L D D B # B L L S D D S B

S

#

B B L S D

0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0 1 1 0 0 0 0 0 0 1 0 0 0 1 0 1 0 0 0 1 1 0 1 0 0 1 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0

D10 72S D9S D9S 73S 72S 72S 100 72S D9S D10 73S 757 D10 72S 320 320 72S 72S 72S 72S D9S D9S D9S D9S 73S 72S 72S 72S 734 757 D10 72S 757 757 72S S80 D10 767 757 320 D10 757 757 733 72S D9S 72S D9S 72S D9S M80 D9S 73S 72S 72S 72S 72S 73S 72S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (27 of 67) [19/1/2003 1:40:22]

F F F F FN FN F F F F F F F F F F F A F F FN F F F F FN F F F F F F F F F F F F F F F F F FN Y F F F F F F F F Y F FN F F A F

Y Y Y Y YN YN Y Y Y Y Y Y Y Y Y Y Y Y Y Y YN Y Y Y Y YN Y Y Y Y Y Y Y Y Y A Y Y Y Y Y Y Y YN K Y Y Y Y Y Y Y Y K Y YN Y Y Y Y

B M V B M Q B M H B M H BN HN QN BN M H B M H B H Q B M H B M H B M H B H Q B M H B M H B Q M B M H B M H Q H K B M Q B M H BN M H B M H B M H B M H B M H BN HN QN B H M B H M B H M B H Q B M H B M Q B M H B M H B M H Y Q H H B M B M H H B M B M H B M H B M Q B M H BN M H L B Q B Q M B M H B Q M B M H B Q M B M H B Q M B M H L B Q B Q M BN M H B M H B M H Q H K B M H

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

NW 486 WN 592 WN 866 WN 30 WN 432 WN 590 DL 408 DL 566 DL 842 WN 214 WN 103 WN 767 WN1615 WN 649 WN 185 UA 317 UA 787 AA 193 AA1412 DL 390 AA 272 DL 629 AA1423 AA1570 DL1404 AA1233 DL 998 AA1234 AA 193 AA1275 DL1023 AA 340 DL1246 AA1080 TW 805 WN 682 NW 270 WN 304 NW 272 WN1615 NW 274 TW 130 WN 631 WN 659 WN 709 WN 590 WN 453 AA1047 AA 386 AA 979 NW1029 NW 472 NW1023 WN 475 WN 435 4X 371 WN 103 4X 214 WN 7 WN 567 WN 609

DTW HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU

845P 650A 910A 130P 450P 605P 940A 1140A 305P 755A 845A 1145A 230P 540P 700P 639A 250P 434P 615A 615A 725A 800A 855A 1100A 1120A 1242P 255P 310P 434P 600P 640P 718P 840P 935P 955P 825A 910A 1005A 1225P 230P 415P 1230P 740A 305P 410P 605P 840P 700A 1155A 545P 800A 1140A 330P 630A 650A 835A 845A 900A 940A 1030A 1105A

TPA ABQ ABQ ABQ ABQ ABQ ATL ATL ATL BNA BNA BNA BNA BNA BNA DEN DEN DEN DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DTW DTW DTW DTW DTW DTW JFK LAX LAX LAX LAX LAX MIA MIA MIA MSP MSP MSP MSY MSY MSY MSY MSY MSY MSY MSY

1128P 750A 1130A 355P 630P 705P 1235P 233P 602P 940A 1155A 135P 550P 730P 840P 810A 417P 740P 718A 724A 837A 903A 1007A 1207P 1225P 151P 355P 423P 550P 712P 742P 825P 944P 1036P 1055P 235P 1243P 425P 400P 935P 751P 450P 1015A 510P 625P 825P 1045P 1030A 328P 917P 1042A 529P 616P 730A 745A 1030A 940A 1240P 1040A 1125A 1205P

S L

B S D # S # S

B L D L

B L D B S S

0 0 2 2 1 0 0 0 0 0 2 0 2 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 3 0 4 0 0 2 1 1 1 1 0 0 0 0 2 0 0 0 1 0 3 0 0 0

72S 733 733 73S 73S 733 M80 D9S M80 733 733 73S 733 733 73S 737 737 S80 S80 73S S80 D9S S80 72S 73S S80 D9S S80 S80 72S M80 S80 73S 72S 72S 73S D9S 733 D9S 733 D9S 72S 73S 735 73S 733 73S 72S 72S 72S D9S D9S D9S 735 73S BEC 733 BEC 73S 73S 733

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (28 of 67) [19/1/2003 1:40:22]

F Y Y Y Y Y Y F F Y Y Y Y Y Y F F F F F F F F F F F F F F F F F FN FN FN Y FN Y FN Y F F Y Y Y Y Y F F F F F F Y Y Y Y Y Y Y Y

Y B M H K L B Q K L B Q K L B Q K L B Q K L B Q B M Q H Y B M Q Y B M Q K L B Q K L B Q K L B Q K L B Q K L B Q K L B Q Y B M Q Y B M Q Y B M H Y B H M Y B M Q Y B H M Y B M Q Y B H M Y B H M Y B M Q Y B H M Y B M Q Y B H Y B H Y B H M Y B M Q Y B H M YN BN MN QN YN B H M YN B Q M K L B Q YN BN M H K L B Q YN BN M H K L B Q Y B M H Y B Q M K L B Q K L B Q K L B Q K L B Q K L B Q Y B H M Y B H M Y B H M Y B M H Y B M H Y B M H K L B Q K L B Q B Q M K L B Q B Q M K L B Q K L B Q K L B Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

WN 123 WN 21 4X 373 WN 308 WN1615 4X 216 WN 625 WN 37 WN 462 4X 375 WN 49 WN 53 WN 57 UA1242 NW 472 UA1482 UA1440 DL 690 TW 344 DL 842 WN 760 WN 949 WN 588 WN 866 WN1741 WN 736 WN 911 WN 421 WN 721 WN 709 WN1595 WN 453 TW 598 WN 592 WN 588 WN 653 WN 721 WN 353 WN 761 TW 344 WN 770 WN 515 TW 598 WN 619 TW 306 WN1633 TW 512 WN 942 WN 728 TW 845 DL1917 AA 715 TW 831 TW 717 TW 44 TW 712 DL1866 NW 34 TW 634 TW 391

HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK

1155A 1240P 115P 120P 230P 230P 340P 455P 615P 620P 740P 845P 940P 800A 1140A 142P 440P 450P 754A 305P 600A 805A 845A 910A 1045A 1110A 1200N 1215P 235P 410P 500P 840P 1030A 650A 845A 1230P 235P 855P 650A 754A 835A 925A 1030A 1145A 225P 405P 528P 630P 915P 335P 455P 615P 730P 420P 650A 405P 525P 600P 905P 1114A

MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY ORD ORD ORD ORD ORD PHL PHL PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX RDU SFO SFO SFO SFO SFO STL STL STL STL STL STL STL STL STL STL STL ABQ ATL ATL ATL BNA BOS BOS BOS BOS BOS DCA

1255P 140P 310P 220P 330P 610P 440P 600P 715P 815P 840P 945P 1035P 1024A 320P 415P 715P 914P 151P 858P 915A 940A 1020A 100P 100P 205P 250P 200P 455P 545P 720P 1015P 406P 950A 1145A 320P 715P 1130P 845A 948A 1030A 1120A 1230P 135P 419P 600P 727P 825P 1105P 825P 716P 1028P 1012P 707P 754A 517P 647P 719P 1009P 1230P

B S S D D BS D

#

B

S

D

SD D D D S

0 0 1 0 0 3 0 0 0 1 0 0 0 0 1 0 0 1 1 1 3 0 0 3 1 2 2 0 1 0 1 0 1 1 1 1 3 1 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 1 0 0 0 0 0 0

73S 73S BEC 73S 733 BEC 73S 735 73S BEC 733 73S 733 73S D9S 733 737 D9S 72S M80 735 733 733 733 735 733 73S 73S 73S 73S 733 73S D9S 733 733 735 73S 73S 735 72S 73S 733 D9S 73S 72S 733 D9S 735 73S L10 310 72S 72S M80 L10 72S 72S D9S M80 M80

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (29 of 67) [19/1/2003 1:40:22]

Y Y Y Y Y Y Y Y Y Y Y Y Y F F F F F F F Y Y Y Y Y Y Y Y Y Y Y Y F Y Y Y Y Y Y F Y Y F Y F Y F Y Y F F F F F F F F F F F

K K B K K B K K K B K K K Y Y Y Y Y Y Y K K K K K K K K K K K K Y K K K K K K Y K K Y K Y K Y K K C C Y C Y C Y Y Y Y Y

L L Q L L Q L L L Q L L L B B B B B B B L L L L L L L L L L L L Q L L L L L L B L L Q L B L B L L Y Y M Y B Y B B B B B

B B M B B

Q Q

B B B M B B B M M M M M Q M B B B B B B B B B B B B V B B B B B B Q B B V B Q B Q B B B B B B Q B Q M M Q Q

Q Q Q

Q Q

Q Q Q Q H Q Q Q M Q Q Q Q Q Q Q Q Q Q Q Q Q K Q Q Q Q Q Q M Q Q K Q M Q M Q Q Q M H Q M Q M Q H M M

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

TW 703 DL1863 TW 873 TW 743 DL1871 AA1387 TW 805 TW 817 TW 805 UA 7 TW 15 AA 1 UA 17 AA 19 MG 300 UA 5 AA 3 TW 815 AA 9 DL 35 AA 21 QF 302 TW 849 UA 15 MG 500 AA 29 TW 3 PA 403 AA1291 AA 903 PA 427 AA 99 PA2453 DL1883 TW 5 FF 37 TW 401 TW 881 TW 829 UA 903 TW 97 TW 711 HP 705 HP 9 HP 7 TW 731 HP 711 AA1161 US1651 AA 715 UA 37 TW 701 AA 59 UA 23 TW 45 UA 25 AA 17 DL 111 AA 65 UA 29 TW 843

JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK

430P 530P 535P 550P 545P 559P 615P 559P 615P 800A 855A 900A 1000A 1030A 1030A 1150A 1200N 415P 430P 535P 600P 600P 600P 620P 640P 730P 759P 715A 715A 1229P 245P 544P 545P 545P 630P 845P 625P 550P 615P 725P 214P 550P 505A 950A 520P 615P 855P 740A 600P 615P 900A 610P 800A 815A 850A 1000A 1200N 540P 545P 545P 620P

DCA DCA DCA DEN DFW DFW DFW DTW HOU LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX MIA MIA MIA MIA MIA MIA MIA MIA MIA MSP MSY ORD ORD PHL PHL PHX PHX PHX PHX PHX RDU RDU RDU SEA SEA SFO SFO SFO SFO SFO SFO SFO SFO SFO

554P 659P 703P 839P 843P 912P 1055P 809P 929P 1049A 1149A 1139A 117P 114P 120P 254P 232P 722P 722P 843P 855P 855P 911P 938P 940P 1021P 1112P 1020A 1005A 327P 557P 900P 906P 858P 941P 1140P 1004P 822P 810P 903P 304P 654P 842A 128P 857P 947P 230A 921A 858P 814P 1209P 920P 1133A 1140A 1208P 127P 321P 852P 920P 902P 948P

S D D D D D D BS B R L L L L L D D D D D D D D D D B R L S D D D D D D D D

B L D D S B # D BS D R BS B L L D D D D

0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0

M80 72S M80 M80 72S 72S 72S 72S 72S D10 L10 D10 767 D10 D8S 767 D10 747 D10 310 D10 D10 L10 763 D8S D10 L10 72S AB3 AB3 72S AB3 AB3 72S 72S 747 72S 72S 72S 727 767 72S 320 757 757 L10 757 72S 734 72S 767 L10 767 767 L10 767 767 310 767 D10 L10

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (30 of 67) [19/1/2003 1:40:22]

F Y B F Y B F Y B F Y B F Y B F Y B Y B Q F Y B Y B Q F Y B F C Y F C Y F Y B F Y B F Y C F Y B P C Y F C Y P Y B C Y B P C Y F J Y F C Y F C Y F Y C F Y B F C Y F Y B F Y H F Y H F Y B F Y H F Y B F Y B F Y M C Q F Y B F Y B F C Y F C Y F C Y F Y B FN YN BN F Y B F Y B F C Y FN YN BN F Y B F Y B F Y B Y B M F C Y F Y B F Y B F C Y F Y B F Y B F C Y F C Y F Y B F C Y

Q M Q Q M H M Q M M B B M H B M B B H M B

M Q M M Q M V M V Q Q H Q M M Q H Q M Q H

B B B H B M B B M B M M T

Q M M M Q Q M M Q M Q Q

Q M Q M B Q B M B Q Q M H KN H K H K B Q H KN M H H Q M H Q H B Q H M M Q B Q M Q H M B M B H M Q B Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AA 15 TW 203 TW 25 TW 391 TW 97 TW 845 TW 717 TW 703 TW 711 TW 77 TW 305 WN 535 WN 533 WN 537 WN 539 WN 801 WN1476 DL 182 DL 264 DL1434 DL1420 DL 178 HP3008 DL 146 AA 380 DL 130 DL 188 AA 350 AA1192 WN 603 AA1138 WN 520 DL 996 NW 932 DL1776 UA 100 NW 330 AA 94 UA 174 DL1420 NW 42 UA 354 NW 554 TW 184 QF 307 AA 12 UA 890 US 608 DL1026 AA 350 NW 302 AA 184 US1492 DL 178 US 34 NW1210 US 626 NW 338 CO1218 DL1522

JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX

915P 627A 1105A 1114A 214P 335P 420P 430P 550P 559P 550P 810A 1110A 215P 535P 755P 800P 1205A 700A 835A 845A 1040A 1205P 1235P 311P 315P 1055P 150A 700A 700A 1235P 200P 1230A 215A 600A 700A 710A 800A 800A 845A 1035A 1122A 100P 120P 200P 200P 210P 855P 1100P 150A 700A 745A 845A 1040A 1145A 1225P 1045P 1100P 620A 620A

SFO STL STL STL STL STL STL STL STL STL TPA ABQ ABQ ABQ ABQ ABQ ABQ ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL BNA BNA BNA BNA BNA BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS DCA DCA DCA DCA DCA DCA DCA DCA DCA DEN DEN

1225A 817A 217P 237P 529P 527P 840P 827P 845P 759P 852P 1055A 150P 455P 820P 1130P 1045P 709A 216P 726P 356P 548P 841P 750P 1234A 1009P 547A 749A 100P 155P 641P 850P 1146A 1215P 405P 520P 503P 425P 421P 740P 703P 918P 1107P 1125P 1016P 1016P 1029P 817A 845A 1146A 425P 533P 546P 835P 900P 935P 809A 846A 943A 140P

S B # # S S S D D D D

S B # B L L L D D S S R L B S BL BS B R BS BD LS LD L LS L L L S SB S BL # B LD LD LD S SB B S

0 0 1 1 1 0 2 1 1 0 0 0 0 0 0 1 0 0 0 2 0 0 1 0 1 0 0 0 0 1 0 1 2 1 1 1 1 0 0 1 0 1 1 1 0 0 0 2 1 1 1 1 1 1 1 1 1 1 0 3

D10 M80 72S M80 767 L10 M80 M80 72S 72S 72S 73S 73S 73S 73S 73S 733 763 767 72S L10 757 320 763 S80 L10 L10 757 757 733 757 733 72S 320 757 D10 320 D10 767 L10 757 757 757 L10 D10 D10 767 733 757 757 757 S80 733 757 734 72S 734 757 AB3 72S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (31 of 67) [19/1/2003 1:40:23]

FN F F F F F F F F Y F Y Y Y Y Y Y FN F F F F F F F F FN FN F Y F Y F F F F F F F F F Y F F F F F YN F@ FN F F F F F F FN F A F

Y Y Y Y C C Y Y Y B Y K K K K K K YN Y Y Y Y Y Y Y Y YN YN YN K Y K Y Y Y Y Y Y Y Y C B Y C J Y C BN Y@ YN Y Y Y Y Y Y YN Y Y Y

B B B B Y Y B B B Q B L L L L L L BN B B B B B B B B BN B B L B L B B B B B B B B Y M B Y Y B Y HN B@ B B B B B B B BN B Q B

H M Q M Q M Q M B Q B Q Q M Q M Q M M V Q M B Q B Q B Q B Q B Q B Q MN QN M Q M Q M Q M Q H K M Q H M M Q MN QN M H M H B Q M H B Q M Q M H M Q M Q M H H M M Q M Q B M Q H M H B Q H M B M QN MN M@ Q@ H Q M H H Q H Q M Q H Q M H HN QN M H H K M Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

UA 270 UA 422 CO1040 UA 354 CO 272 UA 238 CO 598 CO 202 UA 242 UA 646 UA 526 DL 996 AA 814 AA 556 DL1566 AA 606 AA1502 AA1562 DL1684 DL 748 AA 48 AA 480 DL 252 AA 720 DL 720 DL 476 AA 428 DL1442 AA 306 DL 138 AA1226 NW 932 NW 330 NW 332 NW 50 NW 308 NW 338 WN 123 WN 845 WN 611 WN 410 WN 967 WN 863 WN 539 TW 840 DL 34 AA 2 UA 6 MG 200 TW 814 AA 40 AA 4 UA 8 QF 301 AA 32 UA 58 MG 600 AA 22 TW 8 UA 28 TW 702

LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX

630A 805A 840A 1122A 1215P 240P 330P 450P 505P 605P 722P 1230A 1240A 210A 700A 700A 825A 950A 955A 1040A 1116A 1235P 135P 145P 145P 215P 340P 340P 525P 555P 750P 215A 710A 820A 1230P 230P 1100P 530A 735A 800A 1205P 125P 430P 535P 800A 805A 830A 840A 845A 930A 1000A 1200N 1230P 100P 100P 300P 330P 330P 400P 955P 1000P

DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DTW DTW DTW DTW DTW DTW HOU HOU HOU HOU HOU HOU HOU JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK

940A 1131A 1204P 235P 330P 552P 645P 800P 810P 919P 1034P 514A 535A 708A 1154A 1157A 127P 251P 354P 328P 417P 540P 748P 653P 734P 718P 835P 946P 1016P 1040P 1240A 936A 228P 340P 745P 1129P 610A 1135A 200P 210P 645P 715P 1015P 1135P 414P 359P 446P 457P 500P 542P 601P 759P 839P 900P 900P 1106P 1145P 1123P 1155P 549A 600A

B B B L L S S D D D D

B R R L SL L L L D L D L D D D D # S B B L D S

B B R BS B B L L L L L D D D D S S

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 1 0 0 1 0 0 0 0 0 0 0 1 0 1 2 1 2 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

D10 737 AB3 757 733 D10 M80 M80 72S 72S 735 72S D10 S80 L10 D10 D10 S80 73S L10 D10 S80 757 D10 72S 763 D10 733 D10 L10 S80 320 320 D10 D10 D10 757 73S 73S 733 733 733 733 73S 747 310 D10 763 D8S L10 D10 D10 767 D10 D10 767 D8S D10 L10 D10 L10

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (32 of 67) [19/1/2003 1:40:23]

F Y B F Y B A Y Q Y B M Y Q H F Y B A Y Q A Y Q F Y B F Y B F Y B F Y B FN YN B FN YN B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B FN YN BN F Y B F C Y F Y B F Y B Y K L Y K L Y K L Y K L Y K L Y K L Y K L F C Y C Y B F C Y F C Y F Y C F C Y F Y B P Y B F Y B F J Y P C Y F Y B F Y C P C Y F C Y F Y B F C Y

M M H Q K M H H M M M M H H M H H H M M H H M H M M H M H M H M M M B M M B B B B B B B B M B B B B H H M

Q Q K H B Q K K Q Q Q Q M M Q M M M Q Q M M Q M Q Q M Q M Q M H H H M H H Q Q Q Q Q Q Q Q Q H M M Q M M Q

B M B B B M B

H Q M H Q Q Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AA 10 AA 400 PA 445 PA 441 AA1440 AA 200 PA 447 NW 300 NW 302 NW 304 NW 44 NW 308 NW 310 WN 123 CO 230 DL 170 UA 334 AA1456 UA 100 AA 184 UA 708 AA1502 CO 174 UA 106 AA 84 UA 942 AA1200 UA 816 UA 152 QF 305 AA 88 UA 842 UA 120 UA 238 UA 104 AA 380 CO 598 AA1292 UA 502 UA 848 UA 242 UA 202 UA 118 AA 112 UA 270 US1060 UA 48 CO1502 DL 252 UA 98 WN 123 HP1404 WN 603 DL1196 HP 252 WN 611 WN 623 HP1208 WN 617 WN 924

LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX

1000P 730A 915A 1230P 110P 1040P 1045P 130A 700A 810A 1125A 230P 415P 530A 715A 1215P 140P 630A 700A 745A 800A 825A 850A 900A 930A 1000A 1009A 1100A 1200N 1245P 1245P 100P 200P 240P 300P 311P 330P 400P 400P 500P 505P 600P 1130P 1144P 630A 745A 745A 1005A 135P 1045P 530A 600A 700A 705A 735A 800A 825A 850A 905A 1015A

JFK MIA MIA MIA MIA MIA MIA MSP MSP MSP MSP MSP MSP MSY MSY MSY MSY ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD PHL PHL PHL PHL PHL PHL PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX

556A 330P 519P 806P 915P 625A 631A 655A 1225P 139P 456P 750P 936P 1255P 227P 553P 709P 1220P 1257P 147P 201P 434P 522P 259P 320P 353P 414P 450P 556P 642P 642P 700P 752P 1004P 847P 915P 1046P 957P 941P 1041P 1157P 1144P 508A 533A 356P 331P 339P 734P 1245A 632A 740A 813A 910A 917A 950A 1015A 1040A 1105A 1115A 1225P

S R B L L S S S B B L D D B L L # B # B # BS B R L L L L L L L D SD D D S D D D # D # S BL B BS # DS S

#

0 0 0 0 0 0 0 0 0 0 0 0 0 2 1 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 1 0 0 0 1 0 0 1 2 0 0 0 0 0 0 0 0 0 0 0

D10 757 AB3 747 757 D10 AB3 D10 757 72S D10 D10 757 73S AB3 72S 72S D10 D10 S80 767 D10 733 763 D10 D10 S80 72S 72S D10 D10 D10 72S D10 72S S80 M80 S80 72S D10 72S 757 D10 D10 D10 767 757 733 757 757 73S 733 733 73S 73S 733 733 73S 733 733

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (33 of 67) [19/1/2003 1:40:23]

F Y F F Y FN F FN F F C F FN Y A F F F F F F F Y F F F F F F F F F F F F F A F F F F F FN F F F F Y F@ FN Y Y Y F Y Y Y Y Y Y

Y B H M B H M Q Y B M Q C Y B M B H M Q YN B H Y B M Q YN BN M H Y B M H Y B M H Y B M H Y B M H YN BN M H K L B Q Y Q H K Y B M Q Y B M Q Y B H M Y B M Q Y B H M Y B M Q Y B H M H K B V C Y B M Y B H M C Y B M Y B H M C Y B M Y B M Q J Y Y B H M Y B M Q Y B M Q Y B M Q Y B M Q Y B H M Y Q H K Y B H M Y B M Q Y B M Q Y B M Q Y B M Q YN B M Q Y B H M Y B M Q Y B H Q Y B M Q Q H K B Y@ B@ M@ Q@ YN B M Q K L B Q B H K Q K L B Q Y B M Q B H K Q K L B Q K L B Q B H K Q K L B Q K L B Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

HP 425 WN 952 HP1207 WN 410 HP 223 WN 504 DL 252 WN 520 WN 360 WN 871 HP 892 WN 752 WN 719 DL1710 HP 220 WN 661 WN 725 WN 970 WN 797 WN 801 HP 458 WN 217 WN 252 WN 305 DL1612 WN 300 WN 306 DL1719 UA1902 AS 91 NW1282 AS 261 UA1906 DL1443 AS 203 UA1908 UA1108 UA1912 AS 223 UA1764 AS 213 NW 190 UA1716 UA1922 AS 239 DL 173 AS 263 UA1926 DL1578 AS 25 NW1283 UA1500 AS 27 HP 347 DL 171 DL 143 DL 644 UA1730 UA1100 US2246 DL 744

LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX

1015A 1115A 1125A 1205P 1250P 1255P 135P 200P 230P 310P 320P 440P 530P 545P 545P 550P 615P 645P 720P 755P 820P 825P 855P 945P 959P 1010P 1030P 610A 645A 700A 740A 800A 845A 855A 915A 945A 1100A 1145A 1200N 1245P 220P 300P 330P 345P 410P 455P 520P 545P 555P 600P 615P 715P 810P 900P 959P 1250A 600A 630A 700A 700A 700A

PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SFO SFO SFO SFO SFO SFO

1230P 125P 140P 215P 305P 310P 353P 410P 430P 520P 535P 650P 740P 756P 800P 800P 825P 855P 930P 1005P 1035P 1035P 1100P 1155P 1219A 1215A 1250A 841A 928A 935A 1020A 1040A 1137A 1137A 1150A 1218P 336P 228P 240P 325P 455P 540P 732P 626P 640P 723P 800P 828P 821P 840P 854P 958P 1050P 150A 1232A 159A 706A 747A 825A 813A 817A

S

B B B B B B B B S S L L L L S SD D D D D D D D D D S # S S S S S

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0

73S 73S 320 733 73S 73S 757 733 735 73S 73S 73S 735 72S 73S 735 73S 73S 73S 73S 73S 733 733 73S 72S 73S 733 757 735 72S 72S M80 735 72S M80 735 735 735 M80 735 M80 72S 73S 735 72S 72S M80 735 L10 M80 320 735 M80 733 757 757 73S 737 72S 733 73S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (34 of 67) [19/1/2003 1:40:23]

Y B H K Q Y K L B Q F Y B H K Y K L B Q Y B H Q V Y K L B Q F Y B M Q Y K L B Q Y K L B Q Y K L B Q Y H Q Y K L B Q Y K L B Q F Y B M Q Y H Q Y K L B Q Y K L B Q Y K L B Q Y K L B Q Y K L B Q Y H Q V Y K L B Q Y K L B Q Y K L B Q FN YN BN MN QN Y K L B Q Y K L B Q F Y B M Q F Y B M Q F Y B M H F Y B M H F Y B M H F Y B M Q F Y B M Q F Y B M H F Y B M Q F Y B M Q F Y B M Q F Y B M H F Y B M Q F Y B M H F Y B M H F Y B M Q F Y B M Q F Y B M H F Y B M Q F Y B M H F Y B M Q F Y B M Q F Y B M H F Y B M H F Y B M Q F Y B M H YN BN H KN QN F Y B M Q FN YN BN MN QN F Y B M Q F Y B M Q F Y B M Q F Y B H Q F Y B M Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

UA1700 TW 492 UA1102 US 392 DL 856 UA1702 UA1104 US 342 DL 944 UA 809 UA 845 DL1000 US 262 UA 805 US 203 DL1801 UA1108 UA1708 DL1216 US 144 UA1110 UA1710 US 53 DL1020 UA1112 UA1712 US 199 DL 274 UA1114 UA1714 DL 684 US2198 UA 937 UA1716 DL 442 US 890 UA1118 UA1718 DL 550 US2136 UA1120 PA 442 UA1720 UA1122 DL 638 PA 482 US1014 UA1722 US 187 UA1124 DL 772 DL1470 UA1126 US1273 DL 141 UA1130 US1810 PA 448 US1157 US2141

LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX

730A 745A 800A 800A 800A 830A 900A 900A 900A 930A 1000A 1000A 1000A 1030A 1100A 1100A 1100A 1130A 1200N 1200N 1200N 1230P 100P 100P 100P 130P 200P 200P 200P 230P 300P 300P 300P 330P 400P 400P 400P 430P 500P 500P 500P 515P 530P 600P 600P 600P 600P 630P 700P 700P 700P 800P 800P 800P 900P 900P 900P 930P 1000P 1100P

SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO

857A 907A 927A 913A 917A 955A 1027A 1013A 1018A 1057A 1127A 1114A 1118A 1155A 1213P 1218P 1225P 100P 115P 116P 130P 155P 213P 217P 225P 252P 313P 315P 322P 352P 411P 413P 422P 449P 512P 513P 519P 552P 612P 616P 622P 640P 652P 722P 715P 717P 716P 752P 816P 824P 815P 917P 924P 916P 1017P 1016P 1016P 1048P 1113P 1213A

S S S S S S S S S S S # S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S

#

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

73S L10 D10 733 73S 72S 73S 733 73S 737 735 73S 733 73S 733 73S 735 73S 73S 733 73S 73S 734 73S 73S 733 733 73S 733 73S 73S 733 735 73S 73S 733 73S 735 73S 734 733 AB3 733 735 73S 72S 733 737 734 73S 73S 73S 73S 733 757 72S 733 AB3 733 733

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (35 of 67) [19/1/2003 1:40:23]

F F F F F F F F F Y Y F F Y F F F F F F F F F F F F F F F F F F F F F F Y F F F F F F F F F F F F F F F F F FN F FN F FN FN

Y C Y

B Y B

M B M

Q Q Q

Y Y Y Y Y B B Y Y B Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y B Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y YN Y YN Y YN YN

B B B B B M M B B M B B B B B B B B B B B B B B B B B B B B B B M B B B B B B B B B B B B B B B B B BN B BN B BN BN

M M M H M Q Q M H Q H M M M M H M M H M M M H M M M M H M M M H Q M M H M M M M M M H M H M M M M H MN M HN M HN HN

Q Q Q M Q H H Q Q H M Q Q Q Q Q Q Q Q Q Q Q M Q Q Q Q Q Q Q Q Q H Q Q Q Q Q Q Q Q Q Q Q Q Q H Q Q Q QN Q QN Q QN QN

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

TW 72 HP1404 TW 756 TW 492 WN 535 TW 90 DL1434 TW 720 WN 504 TW 184 TW 30 DL 808 DL 170 US 268 US 610 CO1707 DL 167 DL 194 AA 416 DL 464 DL 340 DL 986 AA 466 DL 834 DL1596 AA1556 PA 426 DL 148 AA 954 DL 574 AA1638 AA 397 AA 675 US1586 AA1576 CO 356 NW1828 DL 258 PA 430 US1971 AA1482 NW 776 DL 492 PA 436 AA 672 PA2440 AA 920 DL 464 AA 414 PA 492 NW1828 PA 424 US1971 AA 448 PA 978 AA 942 UA 255 CO1707 DL 509 AA 147 AA 523

LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA

1250A 600A 700A 745A 810A 835A 835A 1000A 1255P 120P 220P 800A 1215P 1255P 1120P 745A 600A 715A 810A 910A 1050A 1245P 145P 215P 355P 440P 555P 600P 701P 910P 739A 110P 649P 700A 755A 1100A 1220P 1225P 125P 135P 150P 200P 430P 535P 709P 735A 844A 910A 1125A 1215P 1220P 135P 135P 140P 540P 659P 725A 745A 410P 630A 745A

STL STL STL STL STL STL STL STL STL STL STL TPA TPA TPA TPA ABQ ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL BNA BNA BNA BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DEN DEN DEN DFW DFW

610A 108P 1233P 349P 220P 207P 412P 327P 815P 653P 802P 337P 843P 824P 650A 123P 741A 858A 1012A 1058A 1241P 233P 350P 410P 543P 635P 758P 754P 911P 1051P 858A 229P 815P 1123A 1111A 346P 526P 325P 439P 827P 507P 635P 728P 856P 1025P 1000A 1106A 125P 157P 244P 235P 404P 618P 411P 815P 931P 1101A 1115A 755P 830A 951A

S S B L B SL L L L B LS L S B B B B S L L S D D D D B # D # R L LS L # SD S D D D D # B SL L # L S S D D B B SD B B

0 1 0 1 1 0 1 0 2 0 0 0 1 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 1 0 0 3 0 1 0 0 0 0 0 1 0 0 0 0 2 0 0 0 1 1 1 0 0

L10 733 767 L10 73S L10 72S 747 73S L10 M80 757 72S 733 733 733 L10 767 72S 757 757 763 72S 763 72S 72S 72S 757 727 72S 72S 72S 72S M80 72S 72S D9S 72S 72S 734 72S 72S 72S 72S AB3 72S 72S 757 727 72S D9S 72S 734 72S 72S 72S 72S 733 72S 757 757

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (36 of 67) [19/1/2003 1:40:23]

FN CN YN Y B H F C Y F C Y Y K L F C Y F Y B F C Y Y K L F C Y F F F FN Y FN F F F F F F F F F F F F FN F F F F F A F F F F F F F F F F F F F F F F F F F F F Y Y@ F F

B K B B B B M B B B

Q Q Q Q Q Q Q Q Q Q

Y B Y B M Q Y B H Q YN BN HN QN Q H K B YN BN MN QN Y B M Q Y B H M Y B M Q Y B M Q Y B Y B H M Y B M Y B H M Y B M Q Y B M H Y B H M YN BN MN QN Y B H M Y B H Y B H M Y B H Q Y B H M Y Q H K Y B M H Y B M Q Y B M Q Y B H Q Y B H M Y B M H Y B M Q Y B M Q Y B H M Y B M Q Y B H M Y B M Q Y B H M Y B M Q Y B M H Y B M Q Y B H Q Y B H M Y B M Q Y B H Y B M Q Q H K B B@ M@ Q@ H@ Y B H M Y B H M

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

DL 703 DL1197 AA1441 AA1599 PA 422 AA1359 DL 509 AA 745 AA 910 AA1080 AA 926 DL 587 US1264 NW 992 PA 710 NW 996 NW 994 PA 438 NW 998 US 784 AA1422 AA 738 AA1080 AA 926 AA 918 PA2454 PA 428 DL1882 TW 4 AA 902 PA 402 AA 424 AA1449 PA 440 US 199 PA 442 AA 411 CO 105 PA 448 AA1550 NW 5 NW 573 UA 267 DL 739 CO1707 AA 978 PA 434 AA1296 PA 418 AA 926 UA 395 AA1522 PA 700 DL 370 CO 514 UA 303 NW 941 AA 976 PA 484 AA 384

MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA

900A 1215P 1255P 205P 205P 323P 410P 448P 655P 659P 703P 730P 820A 830A 830A 1230P 250P 545P 610P 620P 805A 140P 659P 703P 745A 800A 120P 120P 130P 145P 545P 700P 800A 800A 815A 130P 155P 320P 545P 705P 800A 400P 654P 630A 745A 810A 120P 200P 550P 703P 700A 750A 900A 1125A 1225P 1230P 1255P 140P 140P 324P

DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DTW DTW DTW DTW DTW DTW DTW DTW HOU HOU HOU HOU JFK JFK JFK JFK JFK JFK JFK JFK LAX LAX LAX LAX LAX LAX LAX LAX MSP MSP MSP MSY MSY MSY MSY MSY MSY MSY ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD

1055A 220P 300P 423P 427P 533P 608P 705P 901P 1036P 1033P 928P 1250P 1136A 1150A 333P 548P 905P 910P 1046P 957A 330P 854P 1215A 1028A 1048A 435P 424P 437P 444P 844P 951P 1046A 1040A 1251P 431P 420P 718P 846P 955P 1045A 650P 1115P 855A 834A 924A 232P 318P 702P 817P 901A 1009A 1103A 255P 426P 245P 437P 355P 343P 533P

B L L S S S S D D D D D BS B B L S D D D B S D D R B L L # L D D R B # L L # D D B D D # B B # D D B B B L L L S S S S

0 0 0 0 0 0 0 0 0 1 1 0 1 0 0 0 0 0 0 1 0 0 0 2 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 1 1 0 1 0 0 0

72S 72S D10 72S 72S 72S 72S 72S 757 72S 72S 72S 733 D9S 72S D9S D9S 72S 72S 73S 72S 72S 72S 72S AB3 72S AB3 72S M80 AB3 72S AB3 757 747 733 AB3 D10 AB3 AB3 757 72S 72S 733 72S 733 72S 72S 72S 72S 72S 72S AB3 72S 72S 72S 757 D9S AB3 72S 72S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (37 of 67) [19/1/2003 1:40:23]

F Y B F Y F Y B F Y B F Y B F Y B Y B M F Y B Y B H F Y H F Y H F Y B F Y B F Y B F Y B F Y B F Y B F Y B FN YN BN F Y B F Y B F Y B F Y B F Y H F Y H F Y B F Y B F Y B F Y B Y H B F Y B F Y H Y B H F C Y F Y B F Y B F Y B A Y Q F Y B F Y B F Y B F Y B F Y B F Y B Y Q H F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F A Y F Y B F Y B F Y B F Y B F Y B

M

Q

H H M H Q H M B B M H M M M

M M Q M H M Q M M Q Q H Q H

M

Q

H H H H B B M M M Q M M B M B H M H H M H M M M M K M M M M M M H M M Q M M H M H

Q M M M M M Q Q Q M Q Q M Q M Q Q M K Q M H H Q B H Q H Q H H M Q Q H H H M Q M

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AA1354 UA 267 AA 960 DL 574 US 660 AA 668 US1941 AA 692 US1272 DL1120 AA 992 US 541 AA1441 US1550 AA1382 AA1116 AA 34 AA 27 DL 703 UA 379 US 199 PA 442 AA1539 PA 448 TW 245 TW 633 TW 469 TW 471 TW 157 DL 892 US1586 TW 245 PA 300 US 898 US1078 US 180 AA1656 PA 464 US1930 US 761 US1031 PA 986 US1206 DL 939 DL 945 NW 760 DL 320 NW 562 DL1856 DL1239 DL1854 NW 764 NW 616 TW 552 AA1369 UA 256 NW 706 NW1472 NW 720 NW 722 NW 46

MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP

453P 654P 707P 910P 630A 815A 1255P 205P 310P 540P 707P 740P 1255P 700A 610A 1235P 443P 800A 900A 710A 815A 130P 215P 545P 720A 918A 1220P 255P 630P 610A 700A 720A 725A 840A 1030A 1255P 140P 145P 200P 320P 510P 555P 855P 600A 735A 745A 1055A 100P 115P 300P 415P 610P 840P 1107A 1110A 1200N 120P 605P 750A 1140A 115P

ORD ORD ORD ORD PHL PHL PHL PHL PHL PHL PHL PHL PHX PVD RDU RDU RDU SEA SEA SFO SFO SFO SFO SFO STL STL STL STL STL TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL BNA BNA BNA BNA BNA BOS BOS BOS

717P 923P 933P 1230A 909A 1058A 341P 450P 732P 927P 953P 1022P 517P 1121A 819A 242P 700P 1127A 140P 1027A 313P 640P 522P 1048P 1118A 1112A 212P 447P 824P 700A 756A 917A 830A 936A 1125A 154P 239P 250P 300P 415P 605P 700P 950P 922A 1048A 1113A 211P 430P 559P 852P 729P 938P 1159P 250P 245P 330P 319P 806P 1122A 314P 453P

D D D B B L S SD D D S LS B B L D R BL BS # L D D B B # S D # #

# #

B B B L L LS D D S S S S S B L S

0 0 0 1 0 0 0 0 1 1 0 0 1 1 0 0 0 0 1 0 2 1 0 1 2 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 2 0 0 0 1 1 1 0 0 0 0 0

72S 733 AB3 72S M80 72S 734 72S 734 72S 72S 733 D10 734 72S 72S 72S 757 72S 757 733 AB3 767 AB3 M80 M80 L10 72S M80 72S M80 M80 72S M80 F28 73S 72S 72S 733 F28 73S 72S 73S 73S 72S D9S M80 M80 72S 73S 72S D9S D9S D9S S80 73S D9S D9S 757 D9S D10

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (38 of 67) [19/1/2003 1:40:23]

F F F FN F F F F F F F F F F F F F F F Y F F F F F F F F F F F F F F F F F F F F F F F FN F FN F F F F F F FN F F F F FN F F F

Y Y Y YN Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y B Y Y C Y Y Y C Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y YN Y YN Y Y Y Y Y Y YN K Y Y Y YN Y Y C

B H M B M H B H M BN MN QN B H Q B M H B H Q B M H B H M B M Q B M H B H Q H B M B H Q B M H B M H B M H H B M B M Q M Q H B H M B M Q Y H B B M Q B M B Q V Y Q M B Q M B Q M B M Q B H Q B M B M Q B H Q B H Q B H Q M B Q B M Q B H Q B H Q B H M B M Q B H Q BN MN QN B M Q BN M H B M Q B M H B M Q B M Q B M Q B M H BN M H B B B B B B Y

H M M M M M B

M Q H H H H M

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

NW 352 NW 326 NW 356 NW 312 NW 320 AA 870 NW 560 NW 302 NW 324 NW 322 NW 106 UA 389 CO1147 NW 559 NW 561 CO 627 UA 529 NW 563 NW 565 UA 865 CO 589 AA 313 UA 403 NW 567 TW 525 AA 189 NW 401 AA1685 NW 403 AA 374 NW 405 AA 717 NW 407 AA 313 NW 409 NW 740 NW 742 NW 744 NW 746 NW 750 NW 374 NW 752 NW1070 NW 564 NW 140 NW 756 NW 308 NW1020 TW 639 NW1022 TW 202 NW 933 NW 303 NW 305 AA 717 NW 45 NW 309 NW 311 NW 570 TW 496

MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP

240P 600P 845P 700A 755A 828A 1135A 115P 230P 605P 830P 650A 708A 745A 920A 1010A 1020A 1140A 225P 335P 415P 625P 645P 735P 653A 735A 750A 1125A 105P 202P 225P 440P 600P 625P 740P 600A 700A 820A 1000A 100P 115P 300P 400P 500P 500P 605P 850P 110P 349P 615P 1137A 930A 1135A 235P 440P 615P 745P 1030P 755A 805A

BOS BOS BOS DCA DCA DCA DCA DCA DCA DCA DCA DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW HOU HOU HOU JFK LAX LAX LAX LAX LAX LAX LAX MIA MIA

612P 935P 1212A 1013A 1111A 109P 249P 425P 542P 921P 1139P 757A 819A 850A 1020A 1117A 1123A 1245P 326P 426P 516P 1049P 746P 840P 1123A 952A 1018A 141P 336P 420P 450P 703P 831P 842P 1008P 838A 938A 1058A 1240P 348P 359P 541P 754P 744P 915P 855P 1129P 335P 846P 845P 450P 1120A 128P 429P 907P 811P 940P 1228A 1230P 209P

# D B B VS L L S D B B S S S L S S S D D S # B B L S

D D D S B B B S S S

S S S L D D # B L S DS D D S B #

0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0 0 1 0 1 0 0 0 1 0 0 0 0 1

757 D10 757 D9S 320 S80 M80 757 320 757 320 733 72S M80 D9S 73S 737 M80 D9S 72S M80 S80 737 M80 D9S S80 D9S S80 D9S S80 D9S S80 D9S S80 D9S 320 72S 744 757 D10 757 320 D9S M80 D9S D10 D10 D9S D9S D9S 72S D10 D10 757 S80 320 D10 320 72S 72S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (39 of 67) [19/1/2003 1:40:23]

F F FN FN F F F F F F FN F F FN F F F FN F F F F F F F F F F FN F F F F F F FN F F F F F F F F F F F F F F F Y F F F F F FN F F

Y Y YN YN Y Y Y Y Y Y YN Y A YN Y A Y YN Y Y A Y Y Y Y Y

B B BN BN B B B B B B BN B Y BN B Y B BN B B Y B B B B B

M M M M M H M M M M M M Q M M Q M M

H H H H H Q H H H H H Q H H H H Q H

M Q H M M Q M

Q H M Q

Y B YN BN Y B Y B Y B Y B Y B Y B YN BN Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B B M Y B Y B Y B C Y Y B YN BN Y B Y B

M M M M M M M M M M M M M M M M M M M M M Q M Q

H H H H H

M M H B M M M Q

H H M M H H

M H

H H H H H H H H H H H H H H H M H M

M

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

NW 6 NW 453 AA 374 UA 868 AA1578 NW 158 UA 768 UA 486 NW 122 AA 870 UA 70 NW 124 UA 266 NW 126 ZK 34 UA 765 NW 128 AA1369 UA 256 NW 130 NW 112 NW 134 UA 420 AA 80 UA 992 NW 136 NW 138 UA 590 ZK 38 UA 86 NW 140 AA 824 NW 142 UA 554 UA 338 NW 144 AA1085 UA 350 NW 566 NW 680 NW 710 NW 684 NW 8 NW 196 NW 99 HP 606 NW 101 NW 103 NW 105 HP1258 DL1855 NW 107 NW 109 CO 834 NW 7 NW 151 NW 153 NW 47 NW 157 HP 606 NW 351

MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP

100P 1200N 202P 600A 700A 700A 700A 800A 800A 828A 900A 900A 1000A 1000A 1010A 1100A 1100A 1110A 1200N 1200N 100P 200P 200P 225P 300P 300P 400P 400P 420P 500P 500P 545P 600P 600P 700P 700P 705P 800P 835P 740A 100P 235P 600P 830P 745A 845A 930A 1140A 235P 415P 655P 725P 1030P 300P 925A 1145A 230P 745P 1030P 845A 920A

MIA MSY MSY ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD PHL PHL PHL PHL PHL PHX PHX PHX PHX PHX PHX PHX PHX PHX PVD SEA SEA SEA SEA SEA SFO SFO

530P 405P 643P 709A 816A 814A 811A 912A 920A 945A 1007A 1009A 1109A 1115A 1254P 1207P 1215P 1217P 122P 114P 216P 315P 312P 337P 421P 419P 510P 509P 705P 612P 615P 718P 724P 732P 815P 815P 822P 911P 947P 1101A 420P 556P 920P 1146P 1006A 1107A 1156A 200P 452P 643P 1050P 949P 1247A 807P 1100A 123P 405P 925P 1202A 107P 1124A

L S S # V S # # S V # S

S S S S S

S S S S S S S S

B S S D B B B L S S D D S B L S D S BS B

0 1 1 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 1 0

72S 72S S80 72S S80 757 733 73S 747 S80 727 DC9 733 320 BE1 72S D9S S80 73S D9S 320 M80 733 S80 72S D9S D9S 73S BE1 72S D9S S80 72S 733 733 DC9 S80 73S 72S D9S 757 D9S 757 72S 72S 320 D10 72S 757 733 72S 757 72S 737 747 D10 757 D10 757 320 757

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (40 of 67) [19/1/2003 1:40:23]

F F F F F FN F F F F F F F F Y F F F F F F F F F F F F F Y F F F F F F F F FN F F F F F FN F F F F F Y F@ F FN A F F F F FN F F

Y Y Y Y YN YN Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y M Y Y Y Y Y Y Y Y YN Y Y Y Y C YN Y Y Y Y Y B Y@ Y YN Y C Y Y C YN Y Y

B B B B B B B B B B B B B B

M M H M M M M M M M M M M M

H H M Q H H Q Q H H Q H Q H

B M Q B M H B M H B M Q B M H B M H B M H B M Q B M H B M Q B M H B M H B M Q H Q V B M Q B M H B M H B M H B M Q B M Q B M H B M H B M Q B M H B M H B M H B M H Y B M BN M H B M H B H K B M H B M H B M H H K Q B@ M@ Q@ B M H BN M H Q H K Y B M B M H B M H Y B M BN M H B H K B M H

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

NW 353 NW 355 NW 359 NW 361 TW 525 NW 591 TW 496 TW 552 NW 593 TW 227 NW 595 TW 639 TW 570 NW 597 NW 599 NW 478 DL 583 NW 446 NW 448 CO1707 WN 30 WN 590 DL 448 DL 432 DL 662 DL 300 DL 618 DL 480 DL1229 AA1424 WN 103 AA 756 WN1615 AA 828 CO 546 CO1072 CO1646 CO1639 UA 425 CO1707 UA 207 CO 119 DL 651 AA 382 AA1455 DL 793 DL 307 AA1331 AA 121 DL 139 AA 765 DL1507 AA 926 AA 382 WN 304 NW 294 NW1476 WN1615 4X 370 WN 397

MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY

1130A 235P 740P 1030P 653A 735A 805A 1107A 105P 215P 245P 349P 536P 610P 900P 750A 900A 115P 605P 920A 1210P 440P 645A 825A 1005A 135P 335P 535P 845P 630A 955A 1158A 345P 510P 830A 925A 220P 730A 915A 920A 250P 330P 600A 700A 841A 930A 100P 135P 405P 425P 731P 815P 905P 700A 840A 1135A 1215P 345P 615A 620A

SFO SFO SFO SFO STL STL STL STL STL STL STL STL STL STL STL TPA TPA TPA TPA ABQ ABQ ABQ ATL ATL ATL ATL ATL ATL ATL BNA BNA BNA BNA BNA BOS BOS BOS DEN DEN DEN DEN DEN DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DTW DTW DTW DTW DTW HOU HOU

134P 436P 941P 1227A 832A 905A 940A 1245P 237P 347P 420P 524P 722P 741P 1025P 1158A 235P 521P 1005P 123P 355P 705P 912A 1042A 1221P 356P 554P 755P 1059P 755A 1155A 120P 550P 637P 241P 353P 758P 1114A 1112A 1115A 441P 525P 731A 824A 1014A 1055A 227P 305P 541P 553P 902P 946P 1033P 107P 425P 430P 337P 935P 810A 725A

L S D S # S # S

# S B SL L D #

# # S S # S

B L S S S B L S # # # S

# S S

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 3 1 0 0 0 0 0 0 0 0 1 0 1 0 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 4 1 0 3 1 0

757 D10 D10 757 D9S 72S 72S D9S D9S D9S D9S D9S D9S D9S DC9 72S M80 72S 72S 733 73S 733 767 757 72S M80 757 72S 73S S80 733 S80 733 72S 72S M80 733 AB3 737 733 735 733 757 S80 72S 763 767 S80 S80 757 S80 757 72S S80 733 D9S D9S 733 BEC 733

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (41 of 67) [19/1/2003 1:40:23]

F Y B F Y B F Y B F Y B F Y B FN YN BN F Y B F K F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B Y Q H Y K L Y K L F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B Y K L F Y B Y K L F Y B A Y Q F A Y A Y Q A Y Q F Y B Y Q H F Y B A Y Q F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F YN B F Y M Y K L F Y B F Y B Y K L Y B Q Y K L

M M M M Q M Q

H H H H M H M

M Q M Q Q M M M M M M K B B M M M M M

H M H M M H H H Q H H B Q Q Q Q Q Q Q

M H B H B H H Q H H M K M H M M M M M M M M M M M B B M M B M B

Q M Q M Q M K H K K Q B H K Q H H Q H H H Q H H Q H H Q Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

WN 979 WN 711 4X 201 WN 304 WN 443 WN 380 4X 372 WN 843 WN 30 WN 34 WN 837 4X 213 WN 909 4X 374 WN 590 WN 210 WN 453 WN 359 WN 758 TW 880 UA1477 DL 477 DL 139 WN 590 WN 453 AA 485 PA 417 AA 735 PA 433 AA 787 CO1212 NW 397 AA1052 US1068 NW 454 AA 174 UA 480 UA 76 AA1052 TW 574 UA 275 UA 576 AA 248 DL 864 UA1214 US1454 US 184 US1188 WN 453 NW1476 TW 608 UA 363 DL 307 UA 207 CO 485 NW 862 TW 556 TW 338 TW 574 TW 608 CO1568

MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY MSY

645A 755A 835A 840A 940A 1040A 1100A 1115A 1210P 110P 200P 200P 255P 400P 440P 535P 715P 745P 935P 1214P 814A 900A 425P 440P 715P 720A 730A 1230P 320P 634P 815P 635A 100P 445P 530P 731A 800A 1047A 100P 227P 350P 655P 657P 635A 827A 845A 1210P 315P 715P 1215P 536P 734A 100P 250P 330P 730A 804A 1048A 227P 536P 250P

HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU JFK LAX LAX LAX LAX LAX MIA MIA MIA MIA MIA MIA MSP MSP MSP MSP ORD ORD ORD ORD ORD ORD ORD ORD PHL PHL PHL PHL PHL PHX PVD RDU SFO SFO SFO SFO STL STL STL STL STL TPA

750A 900A 1220P 950A 1045A 1145A 1255P 1215P 115P 215P 305P 540P 400P 555P 545P 640P 815P 850P 1035P 404P 1017A 1059A 920P 825P 1045P 959A 1015A 320P 605P 919P 1102P 1031A 608P 1025P 940P 945A 1017A 101P 317P 626P 603P 914P 915P 1150A 112P 145P 343P 650P 1015P 616P 1114P 1029A 450P 830P 756P 1047A 950A 1240P 415P 727P 519P

# B B D

B # L S D # L S S B B L L # S D D B B # L D S D S S LD D S B # D

0 0 2 0 0 0 1 0 0 0 0 3 0 1 0 0 0 0 0 0 0 0 2 2 2 0 0 0 0 0 0 1 1 1 1 0 0 0 0 1 0 0 0 1 1 1 0 0 1 1 1 0 1 2 1 1 0 0 0 0 0

73S 735 BEC 733 733 73S BEC 73S 73S 73S 73S BEC 733 BEC 733 733 73S 73S 73S 72S 72S 763 757 733 73S 72S 72S 72S 72S 72S 733 72S S80 73S 72S S80 73S 72S S80 M80 72S 73S 72S 72S 73S 73S 73S 100 73S D9S D9S 733 767 735 AB3 D9S D9S D9S M80 D9S 733

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (42 of 67) [19/1/2003 1:40:23]

Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y F F F F Y Y F F F F F A F F F F F F F F F F F F F F F F F Y F F Y F F A F F F F F F

K K

L L

B B

Q Q

K K K B K K K K B K B K K K K K Y Y Y Y K K Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y K Y Y B Y Y Y Y Y Y Y Y A

L L L Q L L L L M L Q L L L L L B B B B L L B B B B B Q B B B B B B B B B B B B B B B B B L B B M B B Q B B B B B Y

B B B M B B B B

Q Q Q

B M B B B B B Q M M M B B M M M M M H M H H M M M M M Q M M M M M H H H B M Q Q M M H M Q Q Q Q Q

Q

Q Q Q Q

Q Q Q Q Q M Q Q H Q Q H Q H Q H K H M Q H H Q Q H M H Q H Q Q M Q Q Q H M H H H K H M M M M H

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

DL 170 AA 359 AA 297 AA1271 DL 401 DL 579 UA 996 AA 262 UA 416 DL 635 DL 375 AA 180 UA 860 DL1235 UA1027 DL 575 UA 410 DL1507 AA 142 UA 954 DL 704 DL 177 DL 981 AA 748 UA 542 AA 380 AA1488 AA1401 UA 692 AA 879 AA1369 UA 256 AA 460 AA 315 AA 353 AA1544 UA 584 UA 284 AA 112 UA 340 AA 354 CO 390 UA 764 AA1012 NW 386 CO 358 UA 758 CO 620 AA 166 UA 100 UA 294 US 745 NW 4 NW 390 AA 278 UA 132 AA 848 UA 678 AA1566 UA 790

MSY ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD

625P 905A 1155A 630P 600A 620A 630A 700A 810A 810A 955A 1014A 1107A 1130A 1214P 120P 130P 310P 414P 450P 450P 650P 755P 755P 755P 950P 630A 733A 805A 1144A 115P 200P 250P 530P 630P 750P 755P 600A 655A 755A 910A 915A 1000A 1025A 1025A 1045A 1114A 1135A 120P 200P 345P 345P 400P 400P 430P 500P 530P 630P 800P 800P

TPA ABQ ABQ ABQ ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BNA BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS

843P 1102A 153P 826P 1108A 903A 920A 947A 1108A 1101A 1242P 104P 155P 214P 631P 407P 422P 600P 710P 743P 744P 939P 1036P 1047P 1046P 1234A 757A 907A 929A 107P 245P 330P 412P 654P 809P 928P 916P 912A 1001A 1105A 1224P 211P 117P 133P 330P 446P 222P 402P 425P 520P 659P 758P 709P 850P 747P 819P 850P 947P 1108P 1111P

S R L D # B B B B B S L L L D L S S D D D D # # # V V S S S S S S

B B B V S L L L L # S D # D S D D D D # #

0 0 0 0 2 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 1 0 1 0 0 0 1 0 1 0 0 0 0 0 0

72S S80 S80 72S 72S 72S 72S 72S 73S 757 72S S80 737 72S 73S 72S 72S 757 S80 72S 757 767 D9S S80 73S S80 S80 100 72S S80 S80 73S S80 72S 100 100 72S 757 D10 D10 S80 M80 733 S80 757 733 D10 72S S80 D10 72S 100 M80 72S S80 D10 S80 733 757 733

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (43 of 67) [19/1/2003 1:40:23]

F F F F F@ F F F F F F F F F Y F F Y F F F F FN FN FN FN FN F F F F F F F F F F F F F F A F F F Y F A F F F F F F F F F F F F

Y B M Q Y B H M Y B H M Y B H M Y@ B@ M@ Q@ Y B M Q Y B M Q Y B M H Y B M Q Y B M Q Y B M Q Y B M H Y B M Q Y B M Q B M Q H Y B M H Y B M Q B M Q H Y B M H Y B M Q Y B M H Y B M Q YN BN MN QN YN B M H YN B M Q YN B M H YN B H M Y B H M Y B M Q Y B H M Y B H M Y B M Q Y B H M Y B H Y B H M Y B H M Y B M Q Y B M Q Y B H M Y B M Q Y B H M Y Q H K Y B M Q Y B H M Y B M H Q H K B Y B M Q Y Q H K Y B H M Y B M Q Y B M Q Y B H Q Y B M H Y B M Y B H M Y B M Q Y B H M Y B M Q YN B H M Y B M Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

UA 600 UA 602 AA 246 UA 604 AA1646 UA 606 UA 608 AA 870 UA 610 UA 612 AA1112 UA 614 UA 616 AA 184 UA 618 AA 562 UA 620 UA 622 AA1666 CO 482 UA 624 UA 626 AA 266 UA 293 CO 821 ZK 402 UA 219 AA 379 CO1735 UA 169 UA 89 AA1115 UA 231 UA 435 ZK 406 AA 705 UA 941 UA 205 CO 831 AA 455 CO1765 UA 243 UA 487 AA1199 UA 229 AA 348 AA 299 AA 535 UA 399 DL 877 AA 205 UA 359 CO 642 DL1125 AA 281 DL 819 AA1561 UA 523 AA 347 AA 389 DL1507

ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD

620A 720A 730A 820A 830A 920A 1020A 1025A 1120A 1220P 115P 120P 220P 250P 320P 414P 420P 520P 530P 544P 620P 720P 815P 612A 640A 705A 820A 900A 927A 930A 931A 1144A 1230P 1231P 100P 255P 314P 315P 344P 614P 615P 615P 715P 925P 1000P 615A 644A 730A 753A 830A 915A 952A 1010A 1014A 1115A 1155A 1220P 1224P 150P 305P 310P

DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DCA DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW

900A 1000A 1011A 1102A 1107A 1156A 1259P 109P 159P 256P 359P 402P 503P 533P 559P 658P 659P 808P 813P 1029P 900P 956P 1102P 743A 815A 1055A 959A 1037A 1111A 1111A 1111A 111P 201P 211P 525P 430P 446P 452P 515P 749P 750P 744P 847P 1057P 1136P 836A 901A 958A 1015A 1044A 1136A 1226P 251P 1227P 135P 205P 246P 255P 418P 533P 946P

B B B B B S L S L L # S S S S D D D D D S B B B R S S S L L L S S S S D D D D # B B B B B V S S B L L L L

S

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 3 0 0 0 0 0 0 0 0 4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 2

72S 72S S80 72S S80 72S 72S S80 72S 72S S80 757 72S S80 72S S80 72S 757 S80 M80 72S 72S 757 D10 73S BE1 733 757 733 D10 72S S80 D10 733 BE1 S80 D10 757 M80 S80 733 D10 D10 S80 733 S80 D10 72S 72S D9S D10 733 72S 72S D10 72S 757 733 757 D10 757

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (44 of 67) [19/1/2003 1:40:23]

F F F F F F F F F F F F F F F F F F F Y F F F F F Y F F A F F F F F Y F F F A F A F F F F F F F F F F F A F F F F F F F Y

Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Q Y Y Y Y A M Y Y Y Y Y Y Y Y M Y C Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y B

B B B B B B B B B B B B B B B B B B B H B B B B Y H B B Q B B B B B H B Y B Q B Q B B B B B B B B B B B Q B B B B B B B M

M M M M M M M M M M M M M M M M M M M K M M M M Q Q M M H M M M M M Q M B M H M H M M M M M M M M M M M H M M M M M M M Q

Q Q H Q H Q Q H Q Q H Q Q H Q H Q Q H B Q Q H Q H V Q H K Q Q H Q Q V H M Q K H K Q Q H Q H H H Q Q H Q K Q H Q H Q H H H

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

DL 521 AA 577 AA1243 AA 579 UA 877 DL 505 AA1037 AA 153 NW 161 UA 998 AA 72 NW 200 UA 568 NW1132 NW 386 AA 300 UA 484 NW 532 NW 170 AA 520 UA 994 NW 390 AA 830 UA 446 NW 174 NW 140 UA 108 AA 212 DL 877 UA 971 UA 311 UA 739 TW 746 UA 150 UA 635 UA 39 AA 181 UA 169 UA 101 UA 305 AA1375 AA 281 DL 819 UA 111 AA 409 UA 107 UA 69 UA 259 UA 115 AA 83 AA 389 DL1483 UA 943 AA 889 UA 103 UA 815 AA 89 QF 306 UA 109 AA 615

ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD

345P 425P 620P 645P 655P 659P 805P 930P 610A 630A 645A 725A 810A 900A 1025A 1030A 1105A 1200N 130P 130P 148P 400P 440P 452P 530P 700P 800P 830P 830A 947A 1214P 657P 1230P 314P 700A 815A 915A 930A 1000A 1001A 1025A 1115A 1155A 1200N 1200N 1230P 100P 200P 300P 300P 305P 310P 400P 430P 500P 600P 630P 630P 700P 820P

DFW DFW DFW DFW DFW DFW DFW DFW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW HOU HOU HOU HOU JFK JFK LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX

600P 653P 851P 913P 921P 918P 1033P 1149P 823A 844A 907A 941A 1020A 1115A 1245P 1241P 112P 215P 346P 351P 406P 610P 655P 708P 747P 915P 1010P 1040P 1255P 1234P 302P 947P 339P 622P 914A 1032A 1110A 136P 1222P 1222P 1252P 337P 508P 214P 158P 250P 320P 420P 516P 504P 746P 906P 616P 704P 727P 821P 835P 835P 922P 1053P

S D D D D D

S S V S S

B L L D # S B B R SL L L L LS LS L L L L L D D D S D D D D D D D S

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 1 2 0 0 0 0 0 0 0 1 3 0 0 0 0 0 0 0 0

72S D10 72S D10 72S D9S 757 S80 DC9 733 767 72S D10 D9S 757 S80 72S D9S 72S 757 D10 72S D10 72S 72S D9S D10 D10 D9S 733 737 73S M80 727 72S D10 D10 D10 D10 72S S80 D10 72S 72S D10 D10 72S 72S 72S D10 D10 72S D10 S80 767 763 D10 D10 D10 S80

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (45 of 67) [19/1/2003 1:40:23]

F Y B F Y B F Y B F Y B F Y B F Y B F Y B FN YN B F Y B F Y B F Y B FN YN BN F Y B F Y B FN YN BN F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F C Y F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F C Y F Y B F C Y F C Y F Y B F J Y F Y B F Y B

M M M M M M M M M M M M M M M M M M M M M M M M M M B M M M M M Q M M M H M M M H Q M M H M M M M H Q M B H B B H

Q H H H Q Q H H H Q V H Q H H V Q H H V Q

M H

Q M

V Q H H M V Q Q Q Q M Q Q Q M Q Q Q M M Q Q M Q Q Q Q M Q M M M M M

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AA 435 UA 117 DL 561 AA 671 UA 394 AA1363 AA 469 AA 915 UA 230 PA 483 UA 390 PA 701 AA1293 NW 119 UA 799 NW 121 ZK 31 UA 353 NW 123 UA 221 AA 241 NW 115 UA 599 NW 165 NW 129 AA1333 UA 501 NW 131 UA1453 NW 133 ZK 35 UA 859 NW 135 AA 81 UA 567 NW 137 UA 285 NW 472 AA1052 UA1503 NW 141 UA 217 NW 143 AA 783 UA 663 NW 985 UA 625 NW 109 AA1295 UA 267 UA 463 AA1015 UA 731 AA1223 DL1507 UA 287 NW 475 AA1027 UA 519 CO 478 UA 176

ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD

930P 1000P 530A 630A 744A 855A 1020A 1140A 200P 210P 444P 510P 530P 600A 655A 700A 700A 755A 800A 855A 855A 900A 955A 1000A 1100A 1155A 1155A 1200N 1255P 100P 114P 155P 200P 255P 255P 300P 355P 400P 435P 455P 500P 555P 600P 625P 655P 700P 755P 800P 925P 1000P 745A 915A 1245P 300P 310P 325P 500P 614P 652P 540A 635A

LAX LAX MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSY MSY MSY MSY MSY MSY MSY MSY MSY PHL PHL

1129P 1227A 1045A 1025A 1142A 1248P 217P 335P 604P 603P 839P 903P 933P 715A 817A 827A 955A 919A 920A 1019A 1017A 1027A 1122A 1127A 1225P 123P 121P 130P 214P 220P 410P 319P 320P 423P 417P 420P 525P 529P 608P 626P 635P 717P 731P 756P 825P 825P 908P 929P 1048P 1115P 1002A 1127A 302P 514P 735P 545P 846P 826P 916P 945A 928A

S S SB B B B L L S S D D D S # S # S # V S

S S S S

S S S S S S

B R L S S S D D S B

0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0 1 0

D10 757 72S AB3 757 72S 72S AB3 733 72S 72S 72S AB3 72S 727 D9S BE1 733 320 72S S80 320 73S 72S DC9 S80 733 320 72S D9S BE1 73S D9S S80 72S 320 733 D9S S80 733 747 73S D9S S80 73S 72S 72S 72S S80 733 72S S80 72S 72S 757 73S D9S 72S 73S 737 D10

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (46 of 67) [19/1/2003 1:40:23]

F FN F@ F Y F F F F F F F Y FN F F Y F FN F F F F F F F F F F F Y F F F F F F F F F F F F F F F F FN FN FN F F F F Y F F F F A F

Y YN

B B

H M

M Q

Y B Y Y Y

B M B B B

H H H H H

M

Y B Y B Y B B H YN BN Y B Y B M H Y B YN BN Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B M H Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B YN BN YN B YN B Y B Y B Y B Y B B M Y B Y B Y B Y B Y Q Y B

M M M M M M M Q M M M M M M M M M M M M M Q M M M M M M M M M M M M M M M M M M M M M M M Q M M M M H M

Q H Q V H Q H V Q H Q H H Q H H H Q H Q H V Q H H Q H Q H H Q H Q H H Q H Q H H Q H H Q H H H H H H K Q

M M M

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AA 160 NW 200 UA 456 AA 128 UA 288 AA 360 AA 394 AS 394 UA 236 UA 858 AA 768 UA 114 UA 274 AA 898 HP 111 HP 1 UA 321 AA 407 UA 925 AA 235 UA 965 HP 6 AA 321 HP 10 AA 179 UA 515 NW 109 AA 157 UA 308 AA 208 UA 686 AA 326 AA 716 UA 140 UA 374 AA1362 AA1273 UA 442 AA1288 UA 404 AA 826 UA 488 AA1460 AA 170 UA 293 UA 755 AA 239 UA 863 AA1343 UA 461 AA 391 UA 155 UA 177 UA 457 AA 127 UA 159 AA 673 UA 657 UA 187 AA 273

ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD

705A 725A 805A 1025A 1110A 1215P 120P 120P 144P 325P 430P 505P 810P 815P 315A 605A 828A 910A 930A 1155A 1225P 110P 305P 340P 630P 659P 800P 925P 620A 645A 1145A 115P 430P 450P 755P 755P 626A 635A 1025A 134P 414P 444P 520P 755P 612A 815A 910A 944A 1144A 1217P 255P 330P 331P 530P 625P 705P 930P 1000P 812A 900A

PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PVD PVD PVD PVD PVD PVD PVD PVD RDU RDU RDU RDU RDU RDU RDU RDU SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SFO SFO

958A 1152A 1057A 113P 159P 303P 405P 405P 443P 615P 726P 802P 1058P 1116P 610A 853A 1114A 1153A 1213P 234P 318P 403P 544P 630P 912P 940P 1247A 1205A 927A 946A 249P 417P 741P 802P 1107P 1058P 921A 925A 117P 423P 707P 738P 812P 1044P 1027A 1035A 1141A 1213P 221P 233P 522P 551P 551P 750P 858P 925P 1207A 1234A 1045A 1140A

B S B S L L # LS S S D D #

B B R L L L L G D D D S B B L # D D # # B B L # D D D B B R L L L L D D D D D # S B R

0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

72S 72S 757 S80 72S S80 72S 72S 757 727 757 D10 72S 757 733 320 757 72S 72S S80 733 320 72S 757 72S 72S 72S S80 733 727 733 727 S80 733 733 S80 72S 73S S80 72S S80 733 72S S80 D10 72S 757 D10 S80 D10 72S D10 72S 72S S80 D10 S80 733 D10 757

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (47 of 67) [19/1/2003 1:40:23]

F F F F F F F K F F F F F F YN F F F F F F F F F F F FN FN F F F F F F F F F F F F F F F F F F F F F F F F F F F F FN FN F F

Y Y Y Y Y Y Y

B B B B B B B

M M M M M M M

H H Q H Q H H

Y B M H Y B M H Y B M H Y B M Q Y B M Q Y B M H BN H KN QN Y B H K Y B M Q Y B M H Y B M Q Y B M H Y B M Q Y B H K Y B M H Y B H K Y B M H Y B M Q YN BN M H YN B M H Y B M Q Y B M H Y B M Q Y B M H Y B M H Y B M Q Y B M Q Y B M H Y B M Q Y B M Q Y B M Q Y B M Q Y B M Q Y B M Q Y B M Q Y B M Q Y B M Q Y B M Q Y B H M C Y B M Y B H M Y B M Q Y B H M Y B M Q Y B M Q Y B M Q Y B H M Y B M Q Y B H M YN B M Q Y B M Q Y B M H

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

UA 807 UA 669 AA 117 UA 189 UA 121 AA 575 UA 817 UA 141 AA1525 UA 517 CO1765 AA 41 UA 135 AA 325 AA 457 UA 719 UA1171 TW 679 UA 991 TW 654 AA1063 UA 909 TW 155 AA1011 UA 715 TW 519 AA1365 UA1007 TW 339 TW 367 AA1413 UA 779 TW 121 AA 445 UA 304 UA 974 AA 315 UA 642 DL 919 US1243 DL1433 UA1235 US 771 DL1473 DL1199 DL 223 US 667 DL1444 DL 389 US 277 AA 601 US 535 AA 249 US1141 AA1507 US 404 US 804 DL 202 US1620 US1650 US1432

ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL

935A 945A 1155A 1245P 1246P 300P 320P 455P 530P 550P 615P 645P 655P 820P 930P 1000P 700A 704A 812A 827A 855A 942A 955A 1200N 1214P 117P 244P 327P 405P 600P 615P 700P 715P 925P 744A 314P 530P 810P 535A 630A 845A 935A 950A 1230P 140P 400P 440P 540P 735P 845P 800A 840A 125P 335P 704P 700A 800A 855A 945A 1050A 1200N

SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL TPA TPA TPA TPA ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL BNA BNA BNA BNA BNA BOS BOS BOS BOS BOS BOS

1210P 1220P 227P 310P 321P 539P 548P 732P 807P 815P 1010P 931P 928P 1114P 1156P 1243A 816A 820A 928A 943A 1012A 1100A 1112A 113P 129P 234P 354P 442P 519P 720P 733P 824P 829P 1039P 1119A 649P 1024P 1143P 730A 841A 1047A 1249P 1207P 232P 350P 605P 655P 746P 937P 1101P 910A 957A 237P 446P 817P 814A 913A 1004A 1059A 1203P 113P

L L L L L G D D D D # D D S S S S # S #

# # # #

# # # # # B S SD S B B S L S L S S S D D S B B # S D S S

#

0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

747 72S 767 D10 757 767 747 757 767 72S 733 767 D10 S80 757 733 72S D9S 72S M80 S80 72S 727 S80 737 D9S S80 733 M80 72S 727 72S M80 S80 72S 757 72S 757 M80 733 757 72S D9S 72S M80 757 73S M80 763 73S S80 D9S S80 D9S S80 733 M80 72S 733 D9S D9S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (48 of 67) [19/1/2003 1:40:23]

C Y B M Q F Y B M Q F Y B M H F Y B M Q F Y B M Q F Y B M H F C Y B M F Y B M Q F Y B M H F Y B M Q A Y Q H K F Y B M H F Y B M Q F Y B M H F Y B M H FN YN B M Q F Y B M Q F Y B Q M F Y B M Q F Y B Q M F Y M H B F Y B M Q F Y B Q M F Y M H B F Y B M Q F Y B Q M F Y M H B F Y B M Q F Y B Q M F Y B Q M F Y M H B F Y B M Q F Y B Q M FN Y M H B F Y B M H F Y B M Q F Y B F Y B M Q FN YN BN MN QN F Y B H Q F Y B M Q F Y B M Q F Y B H Q F Y B M Q F Y B M Q F Y B M Q F Y B H Q F Y B M Q F Y B M Q FN YN BN HN QN F Y B H M F Y B H Q F Y B H M F Y B H Q F Y B H M F Y B H Q F Y B H Q F Y B M Q F Y B H Q F Y B H Q F Y B H Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

US1617 TW 712 US1624 US1630 US 176 US1808 US 665 US1576 AA1281 UA 147 CO1795 NW1219 US1626 UA 205 UA 487 UA 239 CO 467 AA1281 AA 201 DL 217 AA1203 DL1061 AA1483 US 275 DL1495 AA 258 NW1191 US1058 NW1219 NW 205 US1674 NW 482 NW 209 US1599 NW 211 TW 712 TW 738 UA 39 UA 99 AA1375 US 695 UA 239 UA 41 US1149 US 435 AA 969 US 433 AA1377 DL 367 AA 921 US 537 NW 7 NW 777 NW 685 NW 379 NW 689 US 233 DL1199 US 553 UA 39

PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL

150P 236P 340P 430P 505P 640P 740P 855P 610A 850A 855A 940A 1050A 105P 500P 530P 540P 610A 724A 825A 1225P 255P 259P 400P 505P 605P 645A 835A 940A 1245P 255P 415P 545P 600P 710P 236P 540P 610A 730A 833A 850A 530P 550P 600P 710A 714A 945A 104P 410P 630P 740P 700A 900A 1145A 505P 750P 900A 140P 255P 610A

BOS BOS BOS BOS BOS BOS BOS BOS DEN DEN DEN DEN DEN DEN DEN DEN DEN DFW DFW DFW DFW DFW DFW DFW DFW DFW DTW DTW DTW DTW DTW DTW DTW DTW DTW JFK JFK LAX LAX LAX LAX LAX LAX LAX MIA MIA MIA MIA MIA MIA MIA MSP MSP MSP MSP MSP MSY MSY MSY ORD

309P 517P 455P 543P 619P 754P 850P 1013P 1036A 1058A 1114A 143P 236P 452P 847P 739P 745P 833A 1000A 1043A 303P 519P 537P 800P 721P 841P 830A 1019A 1125A 230P 440P 603P 733P 748P 900P 326P 639P 1032A 1022A 1252P 1146A 945P 847P 1035P 1006A 956A 1241P 344P 920P 915P 1038P 845A 1052A 136P 655P 942P 1122A 525P 500P 719A

#

B B B L L S SD D D B B B L S S D D D S S S S S D

B BS BL B # D D B B B # D D S B B L D B S L B

0 1 0 0 0 0 0 0 1 0 0 1 1 1 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 1 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0

734 72S 73S D9S 733 734 72S 73S S80 D10 M80 72S 733 757 D10 757 M80 S80 S80 M80 S80 73S S80 M80 72S S80 757 D9S 72S 72S 73S 72S D9S 733 D10 72S L10 D10 757 S80 767 757 757 767 734 72S 734 72S 72S 72S 73S 72S 757 D9S 757 D9S 73S M80 734 D10

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (49 of 67) [19/1/2003 1:40:23]

F Y B H F Y B Q F Y B H Y B H Q F Y B H F Y B H F Y B H F Y B H F Y B H F Y B M F A Y Q F Y B M F F Y B M F Y B M F Y B M A Y Q H F Y B H F Y B H F Y B M F Y B H F Y B M F Y B H F Y B H F Y B M F Y B H F Y B M F Y B H F Y B M F Y B M F Y B H F Y B M F Y B M F Y B H FN YN BN M F Y B Q F C Y B F Y B M F Y B M F Y B H F Y B M F Y B M F Y B M F Y B H F Y B H F Y B M F Y B H F Y B M F Y B M F Y B M F Y B H FN C YN BN F Y B M F Y B M F Y B M F Y B M F Y B M Y B F Y B H F Y B M

Q M Q M Q Q Q Q Q Q H

Q Q Q K M M Q M Q M Q Q M H Q H H Q H H Q H M Q Q Q M Q Q Q Q H Q H Q H Q M H H H H

Q Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AA 271 CO 279 UA 409 AA1375 AA 229 UA 189 AA 391 NW 205 UA 205 UA1017 AA 331 UA 487 UA 369 AA1453 US 353 CO 467 US 289 US1238 US 153 US 540 US 431 AA 649 US 100 AA 529 US 500 US 163 AA 441 US 351 AA 391 UA 183 UA 189 US2461 US 49 TW 217 US 491 TW 419 NW1159 TW 281 US 449 TW 97 US1603 TW 711 US 801 AA 649 US 744 US1701 US 714 NW 482 US 855 DL 389 US 657 WN 664 HP 200 WN 806 WN 702 HP 202 WN 968 WN 393 HP 207 WN 963 WN 521

PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHL PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX

703A 725A 742A 833A 940A 1030A 1245P 1245P 105P 255P 410P 500P 725P 729P 830A 540P 555P 955A 145P 440P 740P 805A 900A 1159A 300P 550P 650P 855P 1245P 830A 1030A 550P 550P 700A 830A 937A 1220P 1248P 300P 359P 600P 719P 700A 805A 830A 1000A 1225P 415P 430P 735P 850P 700A 810A 825A 925A 1034A 1035A 1135A 107P 150P 230P

ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD PHX PHX PHX PVD PVD PVD PVD RDU RDU RDU RDU RDU RDU RDU SEA SFO SFO SFO SFO STL STL STL STL STL STL STL STL STL TPA TPA TPA TPA TPA TPA TPA TPA TPA ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ

815A 956A 856A 945A 1046A 1140A 154P 325P 218P 402P 523P 612P 840P 845P 1156A 1020P 923P 1057A 241P 538P 838P 928A 1028A 127P 421P 714P 813P 1021P 522P 1129A 310P 900P 900P 832A 1002A 1107A 347P 220P 438P 529P 734P 845P 940A 1158A 1102A 1242P 422P 934P 710P 1145P 1133P 805A 920A 935A 1030A 1144A 1145A 1240P 215P 255P 335P

B S B B V S L S S S S S # D B D D

V S # S S L BS SL D D B B S L # S S D D B V B L LS D D D S

0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1 1 0 1 0 0 0 0 0 0 0 0 0 0 0

757 733 757 S80 757 D10 72S 72S 757 72S S80 D10 727 72S 733 M80 733 73S M80 D9S 733 72S 73S 72S 733 D9S S80 100 72S D10 D10 767 767 72S 73S 72S D9S 72S 100 767 73S 72S 734 72S M80 733 733 72S 734 763 734 733 733 73S 73S 733 73S 73S 73S 733 733

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (50 of 67) [19/1/2003 1:40:23]

F Y B M H F A Y Q H F Y B M Q F Y B M H F Y B M H F Y B M Q F Y B M H F Y B M H F Y B M Q F Y B M Q F Y B M H F Y B M Q F Y B M Q F Y B M H F Y B H Q A Y Q H K F Y B H Q F Y B H Q F Y B H Q Y B H Q M F Y B H Q F Y B H M F Y B H Q F Y B H M F Y B H Q Y B H Q M F Y B H M F Y B H Q F Y B H M F Y B M Q F Y B M Q F Y B H Q F Y B H Q F Y B Q M F Y B H Q F Y B Q M F Y B M H F Y B Q M F Y B H Q F C Y B Q F Y B H Q F Y B Q M F Y B H Q F Y B H Q F Y B H Q F Y B H Q F Y B H Q F Y B M H F Y B H Q F@ Y@ B@ M@ Q@ F Y B H Q Y K L B Q F Y B H K Y K L B Q Y K L B Q Y B H K Q Y K L B Q Y K L B Q Y B H K Q Y K L B Q Y K L B Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

WN 668 HP 205 WN 466 HP 208 WN 998 WN 497 HP 212 WN 938 WN 801 DL 314 UA 410 HP3002 DL 222 DL1898 DL 296 CO 846 HP3004 AA 162 WN 603 AA1222 WN 520 HP 826 NW1268 AA 848 HP 820 DL 360 US 4 CO1182 HP 182 DL 314 US 632 US 356 HP 264 AA 562 UA 622 NW 250 HP 268 NW 106 UA 396 CO 562 HP 497 HP 490 CO 280 HP 499 UA 164 HP 492 CO1170 HP 496 UA 706 CO1182 HP 491 CO1672 HP 493 AA 534 DL 286 AA 316 DL1112 AA 570 HP 546 CO 280

PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX

330P 345P 405P 606P 625P 820P 847P 1010P 1030P 1210A 835A 1024A 1030A 200P 225P 235P 615P 910A 935A 245P 430P 910A 1115A 1225P 103P 225P 235P 510P 1010P 1210A 1225A 825A 840A 1045A 1202P 150P 233P 345P 800A 800A 855A 1040A 1050A 1200N 101P 104P 140P 349P 415P 510P 611P 755P 834P 235A 600A 835A 900A 1010A 1033A 1050A

ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ABQ ATL ATL ATL ATL ATL ATL ATL ATL BNA BNA BNA BNA BOS BOS BOS BOS BOS BOS BOS BOS DCA DCA DCA DCA DCA DCA DCA DCA DCA DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DFW DFW DFW DFW DFW DFW DFW

435P 454P 510P 713P 730P 925P 958P 1115P 1130P 529A 422P 408P 410P 1042P 753P 946P 1202A 100P 155P 638P 850P 357P 711P 850P 748P 1040P 1127P 123A 648A 755A 806A 403P 358P 658P 808P 959P 951P 1139P 940A 944A 1041A 1229P 1233P 338P 240P 248P 319P 533P 559P 649P 758P 932P 1020P 538A 908A 1148A 1217P 129P 200P 407P

S BS L L D D # D R D L LD LD L # # SD # SB S BS S LD LD L S D B #

S S S # S # S S

B B B L S S

0 0 0 0 0 0 0 0 0 0 1 0 0 2 0 1 0 0 0 0 0 0 1 1 0 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1

733 733 733 733 735 73S 733 733 73S 757 72S 320 72S 72S 757 72S 320 S80 733 S80 733 757 72S S80 757 72S 734 733 757 757 733 734 733 S80 757 320 733 320 72S 72S 73S 733 M80 733 72S 733 72S 733 733 733 73S 73S 733 72S 72S D10 72S D10 733 M80

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (51 of 67) [19/1/2003 1:40:23]

Y F Y F Y Y Y Y Y FN F F F F F A F F Y F Y F F F F F F Y FN FN FN F Y F F F F F F A Y Y A Y F F Y Y F Y Y A Y FN FN F F F F A

K Y K Y K K B K K YN Y Y Y Y Y Y Y Y K Y K Y Y Y Y Y Y Q YN YN YN Y B Y Y Y Y Y Y Y B B Y B Y Y Q B Y Q B Y B YN YN Y Y Y Y Y

L B L B L L H L L BN B B B B B Q B H L H L B B B B B B H BN BN BN B H B B B B B B Q H H Q H B B H H B H H Q H B BN B B B B Q

B H B H B B K B B MN M H M M M H H B B B B H M M H M H K H MN HN H K M M M H M M H K K H Q M H K K M K K H K H MN H M H H H

Q K Q Q Q Q Q Q Q QN H K Q Q Q K K M Q M Q K H Q K Q Q B KN QN QN Q Q Q Q H K H Q K Q Q K Q K B Q H B Q K Q M QN M Q M K K

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AA 740 DL 588 HP 540 HP 547 AA1074 DL 252 HP 548 AA1530 NW 256 NW 248 DL 588 NW 250 NW 252 WN 559 WN 123 WN 308 WN 611 WN 920 WN 361 WN 410 WN 882 WN 728 WN 560 WN 674 WN 997 WN 781 HP 290 TW 730 HP 292 HP 790 WN 291 HP 14 DL1467 WN 596 WN1403 WN1602 HP 15 WN 921 HP 17 WN 602 HP 18 WN 665 DL 369 WN 440 HP 20 WN1496 WN 426 WN 778 WN 994 HP 24 WN 947 WN 707 WN 344 HP 29 WN 923 WN 709 DL1793 WN 931 WN1545 HP 31 WN 212

PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX

1250P 1255P 113P 337P 340P 435P 617P 705P 1240A 1000A 1255P 150P 300P 700A 755A 845A 1030A 1100A 100P 230P 350P 530P 635P 650P 750P 850P 900A 930A 120P 845P 630A 650A 650A 700A 725A 750A 800A 830A 925A 930A 1025A 1030A 1105A 1140A 1200N 1245P 120P 200P 230P 230P 400P 435P 455P 455P 535P 605P 640P 640P 710P 730P 740P

DFW DFW DFW DFW DFW DFW DFW DFW DTW DTW DTW DTW DTW HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU JFK JFK JFK JFK LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX

417P 415P 439P 701P 657P 748P 945P 1016P 617A 337P 830P 733P 851P 1145A 1135A 100P 210P 435P 440P 645P 800P 900P 1045P 1100P 1120P 1220A 340P 405P 800P 325A 640A 706A 703A 715A 740A 810A 820A 845A 945A 945A 1045A 1050A 1130A 1200N 1220P 105P 140P 220P 250P 250P 420P 455P 515P 515P 555P 625P 649P 700P 730P 750P 800P

L L S S D D S D S L LD L D

B L L D

#

S

S

0 0 0 0 0 0 0 0 0 0 1 0 0 2 0 1 0 3 0 1 1 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

72S 72S 733 733 S80 757 73S 72S 72S 72S 72S 320 72S 73S 73S 73S 733 733 735 733 733 73S 73S 73S 735 733 757 L10 757 320 733 73S 72S 73S 733 733 73S 733 73S 733 320 73S 757 733 73S 733 733 735 73S 73S 73S 73S 735 73S 73S 73S 73S 73S 73S 73S 733

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (52 of 67) [19/1/2003 1:40:23]

F Y B F Y B F Y B Y B H F Y B F Y B Y B H F Y B FN YN BN F Y B F Y B F Y B F Y B Y K L Y K L Y K L Y K L Y K L Y K L Y K L Y K L Y K L Y K L Y K L Y K L Y K L F Y B F C Y F Y B FN YN BN Y K L Y B H F Y B Y K L Y K L Y K L Y B H Y K L Y B H Y K L F Y B Y K L F Y B Y K L Y B H Y K L Y K L Y K L Y K L Y B H Y K L Y K L Y K L Y H Y K L Y K L F Y B Y K L Y K L Y B H Y K L

H M M Q H K K Q H M M Q K Q H M M H M H M Q M H M H B Q B Q B Q B Q B Q B Q B Q B Q B Q B Q B Q B Q B Q H K B Q H K H KN B Q K Q M Q B Q B Q B Q K Q B Q K Q B Q H K B Q M Q B Q K Q B Q B Q B Q B Q K Q B Q B Q B Q B B M B B K B

Q Q Q Q Q Q Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

WN 757 DL1797 WN1950 WN 274 HP 35 WN 453 NW 100 NW 110 HP 580 NW 102 NW 104 NW 106 NW 108 HP 586 WN 123 WN 308 UA1276 AA 890 HP 2 AA 148 UA 410 HP 4 AA 562 UA 622 AA 848 AA 204 UA 762 HP 8 HP 333 DL 648 US 796 US 431 AA1222 AA1074 DL 252 US 431 HP 523 NW1233 AS 603 HP 844 AS 613 HP 841 AS 605 AS 608 UA1770 HP 845 AS 611 AS 607 HP 840 UA1444 HP 600 WN 640 HP 604 WN 588 UA1040 HP 606 UA1770 WN 487 HP 605 UA1432

PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX

845P 850P 900P 935P 955P 1030P 815A 930A 1035A 105P 230P 345P 600P 613P 755A 845A 1255A 140A 800A 810A 835A 1030A 1045A 1202P 1225P 219P 310P 605P 850P 830A 920A 1245P 245P 340P 435P 1245P 715A 745A 810A 932A 1110A 1210P 105P 235P 235P 457P 530P 605P 940P 710A 740A 810A 947A 1045A 1121A 1205P 235P 440P 500P 505P

LAX LAX LAX LAX LAX LAX MSP MSP MSP MSP MSP MSP MSP MSP MSY MSY ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD PHL PHL PHL PHL PHL PHL PVD SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO

905P 903P 925P 950P 1015P 1045P 1218P 131P 245P 503P 625P 741P 951P 1017P 1255P 220P 503A 555A 1220P 1221P 1245P 244P 310P 415P 443P 647P 720P 1022P 118A 415P 336P 700P 1032P 1141P 1245A 838P 1130A 1125A 1000A 1157A 110P 316P 305P 535P 628P 718P 720P 855P 1158P 810A 839A 905A 1050A 1145A 1230P 107P 335P 540P 602P 605P

B B S L L D D S

# B B B L L L L D D D S B B L D # DS L SB B S L S L L S S D D S B # # L S # S S

0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 2 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 1 1 1 2 1 0 0 0 1 0 1 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0

73S 72S 733 73S 733 73S 757 72S 733 D10 72S 320 757 733 73S 73S 733 S80 320 72S 72S 757 S80 757 S80 S80 72S 320 733 72S 733 733 S80 S80 757 733 733 320 72S 733 M80 733 M80 72S 72S 73S 72S 72S 73S 72S 320 73S 320 733 733 320 72S 73S 733 72S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (53 of 67) [19/1/2003 1:40:23]

Y F Y Y Y Y F F Y F F F FN Y Y Y FN FN F F F F F F F F F F YN F F F F F F@ F Y FN F Y F Y F F F Y F F Y F F Y F Y F F F Y F F

K L B Q Y B M Q K L B Q K L B Q B H K Q K L B Q Y B M H Y B M H B H K Q Y B M H Y B M H Y B M H YN BN M H B H K Q K L B Q K L B Q YN B M Q YN B M H Y B H K Y B M H Y B M H Y B H K Y B M H Y B M Q Y B M H Y B M H Y B M Y B H K BN H KN QN Y B M Q Y B H Q Y B H Q Y B M Q Y B M Q Y@ B@ M@ Q@ Y B H Q B H K Q YN BN M H Y B M H B H K Q Y B M H B H K Q Y B M H Y B M H Y B M Q B H K Q Y B M H Y B M H B H K Q Y B M Q Y B H K K L B Q Y B H K K L B Q Y B M Q Y B H K Y B M Q K L B Q Y B H K Y B M Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

WN 721 WN 753 HP 602 UA1091 WN 417 HP 301 TW 146 WN 696 TW 284 HP1404 TW 824 WN 689 TW 316 WN 510 HP1408 WN 966 TW 260 WN 504 WN 728 HP1406 CO1002 CO1619 AA 637 DL 929 NW 799 DL 679 US1162 US 291 US1755 US1109 US 696 UA 523 NW 695 NW1003 NW 799 NW 695 NW1449 UA1545 UA1185 AA 133 UA 433 AA 669 UA 523 AA 793 UA 643 AA 381 UA1763 US2385 US 531 US 173 US 221 AA 637 US1755 AA 769 AA1559 US 697 US 291 DL 553 US 430 DL1175 DL1143

PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD PVD RDU RDU

520P 715P 745P 748P 915P 1120P 255A 815A 831A 905A 941A 1000A 1218P 1220P 117P 155P 310P 335P 530P 614P 910A 630A 735A 830A 930A 505P 550P 630A 1100A 140P 635P 1005A 355P 630A 930A 355P 655P 640A 1155A 625A 715A 915A 1005A 1230P 400P 712P 755P 640A 1255P 400P 655P 735A 1100A 1130A 616P 730A 630A 750A 600P 615A 800A

SFO SFO SFO SFO SFO SFO STL STL STL STL STL STL STL STL STL STL STL STL STL STL TPA ATL ATL ATL ATL ATL ATL DCA DCA DCA DCA DFW DFW DTW DTW DTW DTW MIA MSY ORD ORD ORD ORD ORD ORD ORD ORD PHL PHL PHL PHL RDU RDU RDU RDU SEA TPA TPA TPA ATL ATL

715P 805P 849P 847P 1010P 140A 646A 100P 1225P 108P 135P 240P 415P 425P 524P 640P 709P 815P 1105P 1011P 441P 1113A 1138A 1104A 223P 736P 940P 752A 1222P 304P 800P 255P 903P 840A 1139A 603P 902P 1143A 352P 753A 858A 1043A 1143A 203P 548P 849P 931P 751A 203P 508P 803P 925A 212P 122P 807P 1232P 1045A 1205P 1047P 740A 915A

D

# B S B L S D

D BS B B B L D S S

# L D S

SB S B B V L L S D # S # # B # L D B SB B DS # #

1 0 0 0 0 1 0 1 0 0 0 1 0 0 0 1 0 1 1 0 1 1 1 0 1 0 1 0 0 0 0 1 1 0 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 1 1 1 0 0

73S 73S 73S 72S 73S 733 72S 733 72S 733 72S 73S 72S 733 733 73S 72S 73S 73S 320 M80 73S 727 M80 D9S M80 D9S M80 D9S 733 D9S 733 D9S D9S D9S D9S D9S 733 73S S80 733 S80 733 727 733 727 733 733 73S 73S M80 727 D9S 72S S80 733 M80 M80 M80 72S 73S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (54 of 67) [19/1/2003 1:40:23]

Y Y Y F Y FN FN Y F Y F Y F Y Y Y F Y Y F A F F F F F F F F F F F F F F F FN F F F F F F F F F F F F F F F Y F F F F F F FN F

K L K L B H Y B K L YN BN YN B K L Y B B H Y B K L Y B K L B H K L Y B K L K L Y B Y Q A Y Y B Y B Y B Y B Y B Y B Y B Y B Y B Y B

B Q B Q K Q M Q B Q H KN Q M B Q Q M K Q Q M B Q Q M B Q K Q B Q Q M B Q B Q H K H K Q H M H M Q M H M Q H Q H Q H Q H Q H Q M Q

Y B M H Y B M H Y B M H YN BN M H Y B M Q Y B M H Y B M H Y B M Q Y B M H Y B M Q Y B M H Y B M Q Y B M H Y B M Q Y B H Q Y B H Q Y B H Q Y B H Q Y B H M B H Q M Y B H M Y B H M Y B H Q Y B H Q Y B M Q Y B H Q YN BN MN QN Y B M Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

DL 495 AA 637 DL 297 AA 467 DL1059 US1636 DL 359 AA 715 AA1159 US 303 AA1571 AA 483 US1072 AA1194 US1630 AA 736 AA1170 US1691 AA 766 US1642 AA 298 US1004 AA1610 AA 429 AA1373 DL 427 AA1277 AA 383 AA1497 DL1193 AA1243 AA1089 AA1527 AA 596 AA1062 DL 412 AA1046 AA 542 US1100 AA1160 AA1551 AA 35 AA1635 AA1219 UA1079 AA 887 UA1479 AA1207 AA1302 AA1243 UA 423 US 205 AA1312 US 270 AA 710 US1708 US1630 AA 358 US 64 AA1258

RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU

940A 1010A 105P 200P 430P 815P 825P 900P 820A 115P 200P 740P 845A 850A 245P 320P 750P 710A 850A 1005A 320P 725P 745P 635A 810A 855A 1005A 100P 325P 410P 420P 504P 655P 900A 325P 640P 740P 330P 420P 740P 1005A 210P 900P 701A 735A 855A 1035A 1258P 320P 420P 508P 625P 740P 750A 900A 1130A 245P 320P 530P 745P

ATL ATL ATL ATL ATL ATL ATL ATL BNA BNA BNA BNA BOS BOS BOS BOS BOS DCA DCA DCA DCA DCA DCA DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DTW DTW DTW DTW JFK JFK JFK MIA MIA MIA ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD PHL PHL PHL PHL PHL PHL PHL

1104A 1138A 227P 328P 550P 1054P 939P 1028P 856A 313P 242P 818P 1023A 1035A 543P 507P 957P 804A 1000A 1059A 414P 821P 850P 829A 1003A 1051A 1200N 300P 535P 611P 851P 713P 903P 1043A 514P 955P 921P 510P 733P 913P 1205P 421P 1103P 808A 843A 1007A 1145A 202P 430P 525P 614P 855P 845P 900A 1022A 1242P 358P 437P 643P 904P

S S

B

B B #

S

# B B B S L S S SD D D B S

D

B B B S L S D S S V S # S

0 0 0 0 0 1 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0

M80 727 M80 S80 72S D9S D9S 72S S80 D9S 727 72S 100 S80 D9S S80 D10 100 S80 D9S S80 73S S80 727 S80 M80 S80 S80 S80 73S 72S 72S S80 72S 72S 757 72S 72S 73S 72S 72S 72S 72S S80 733 72S 73S S80 S80 72S 72S D9S S80 733 72S 73S D9S S80 D9S 72S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (55 of 67) [19/1/2003 1:40:23]

F F F F F Y F FN F Y F F F F Y F F Y F F F F F F F F F F F F F F F F F F F Y Y F F F FN F F F F F F F F F F F F F F F F F

Y Y Y Y Y B Y Y Y B Y Y Y Y B Y Y B Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y B B Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y

B B B B B H B B B H B B B B H B B H B B B B B B B B B B B B H B B B B B B M H B B B B H B H B H H H B B H B B B B B B B

M H M H M Q M H M Q M M H H Q H H Q H H H H H H H M H H H M M H H H H M H Q Q M H H H B M B M B B B M H B H H H H H H H

Q M Q M Q M Q M H M H H Q M M M M M M Q M Q M M M Q M M M Q Q M M M M Q M H M Q M M M M Q M Q M M M Q Q M Q M Q Q M Q M

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

US1691 AA 334 AA1340 US1004 AA 750 TW 321 TW 577 TW 505 DL1175 US1170 AA 649 AA1403 AA1005 UA 509 DL 706 DL 830 DL 304 AA1406 AA 440 NW 46 US1438 TW 150 UA 56 TW 80 CO 758 NW 986 UA 620 DL 166 AA 266 CO1232 UA 228 UA 509 CO1272 UA 246 CO 136 UA 700 UA 146 CO1262 UA 362 CO1214 AA 396 AA1202 DL 174 AA1406 DL 994 AA1096 DL1720 AA1016 DL 558 AA 440 DL 198 AA1442 NW 578 NW 282 NW 580 NW 582 TW 708 UA 34 DL 620 AS 22 NW1280

RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU RDU SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA

710A 900A 330P 725P 745P 701A 117P 434P 615A 815A 1010A 210P 900P 800A 830A 325P 1025P 750A 119P 700A 815A 1004A 100P 112A 635A 705A 955A 1230P 115P 620A 620A 800A 900A 1117A 1150A 1222P 243P 300P 425P 525P 1212A 619A 630A 750A 825A 1036A 1145A 1155A 1235P 119P 145P 815P 845A 1230P 155P 1130P 850A 120P 650A 700A 700A

PVD PVD PVD PVD PVD STL STL STL TPA TPA TPA TPA TPA ABQ ATL ATL ATL BNA BNA BOS BOS BOS BOS DCA DCA DCA DCA DCA DCA DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DTW DTW DTW DTW JFK JFK LAX LAX LAX

950A 1040A 513P 1013P 922P 814A 236P 554P 1000A 1122A 1158A 403P 1048P 134P 349P 1048P 543A 423P 956P 453P 614P 846P 1100P 1040A 522P 453P 659P 945P 1102P 947A 944A 1129A 1226P 240P 319P 352P 604P 629P 752P 841P 545A 1203P 1207P 137P 340P 417P 713P 654P 743P 714P 727P 547A 353P 736P 901P 627A 450P 921P 911A 940A 931A

S B # B # D # S

# B D S B L BS BS LD L SB BS # LD L L B B B B L L L S S D D B B B BL L L L LD L L B L D S B L B B B

1 0 0 1 0 0 0 0 1 1 0 0 0 1 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 1 1 0 0 2 0 0 0 0 0 0 0 0 0

100 72S S80 73S 727 D9S D9S D9S 72S 734 72S 72S 72S 733 L10 72S 763 S80 757 D10 733 M80 757 M80 M80 320 72S 72S 757 AB3 D10 733 M80 D10 72S 735 D10 M80 735 72S 72S S80 72S S80 757 S80 757 S80 L10 757 757 S80 757 757 72S 320 L10 767 757 M80 757

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (56 of 67) [19/1/2003 1:40:23]

Y F F F F F F F F@ F F F FN F F F FN F F F Y F F FN A F F F F F F F A F A F F A F A FN F F F F F F F F F F F F F F F F F F F F

B H Q M Y B H M Y B H M Y B H Q Y B H M Y B Q M Y B Q M Y B Q M Y@ B@ M@ Q@ Y B H Q Y B H M Y B H M Y B H M Y B M Q Y B M H Y B M Q YN BN MN QN Y B H M Y B H M C Y B M B H Q M Y B Q M Y B M Q YN B Q M Y Q H K Y B M H Y B M Q Y B M Q Y B M H A Y Q H Y B M Q Y B M Q Y Q H K Y B M Q Y Q H K Y B M Q Y B M Q Y Q H K Y B M Q Y Q H K YN B H M Y B H M Y B M Q Y B H M Y B M Q Y B H M Y B M Q Y B M Q Y B M Q Y B H M Y B M Q Y B M Q Y B M H Y B M H Y B M H Y B M H C Y B Q Y B M Q Y B M Q Y B M H Y B M H

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

UA1901 AS 210 DL 170 UA 112 AS 214 AS 228 UA1909 NW 189 AS 234 DL1708 AS 238 UA1915 UA1919 DL1710 UA1247 AS 226 UA1921 DL 192 AS 96 UA1925 DL 887 AS 202 NW 191 UA1929 DL1714 DL 174 DL 558 AA 26 NW 158 NW 46 NW 8 NW 156 NW 154 DL 170 DL1720 AA 632 NW 158 UA 228 TW 326 AS 394 AA 394 UA 154 UA 880 AA 716 UA 240 UA 620 UA 246 CO 136 AA 266 UA 862 AA1692 UA 156 UA 148 AS 394 AA 394 AS 602 HP1492 AS 612 HP1299 AS 610

SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA

700A 800A 830A 830A 1015A 1100A 1130A 1130A 1200N 1235P 100P 100P 230P 230P 300P 345P 400P 425P 500P 530P 545P 630P 630P 700P 755P 630A 1235P 1255P 100A 700A 1140A 235P 455P 830A 1145A 1201A 100A 620A 644A 650A 650A 705A 706A 940A 954A 955A 1117A 1150A 115P 135P 335P 630P 1135P 650A 650A 645A 715A 1015A 1220P 140P

LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX MIA MIA MIA MSP MSP MSP MSP MSP MSY MSY ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD PHL PHL PHX PHX PHX PHX PHX

930A 1035A 1052A 1100A 1250P 135P 157P 158P 235P 252P 330P 327P 456P 448P 714P 615P 627P 843P 730P 758P 909P 905P 900P 932P 1016P 445P 1215A 930P 605A 1210P 446P 740P 1000P 553P 950P 541A 814A 210P 248P 1228P 1228P 1245P 1245P 320P 331P 331P 703P 739P 717P 718P 915P 1213A 508A 405P 405P 1030A 1105A 155P 412P 525P

B B B B L L L L L L L L S S S D D S D D D D D D D BL LD L S B L D D BL L S BL # B B B B L L L LD LD L L D D # BL # B S L S L

0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 1 0 0 0 0 1 2 0 0 0 0 0 0 1 2 0 1 1 1 0 0 0 0 0 0 0 1 1 0 0 0 0 0 1 1 0 0 0 0 0

735 M80 72S 735 M80 M80 735 72S M80 72S 72S 735 735 72S 735 M80 735 757 72S 735 72S M80 72S 735 757 72S L10 757 757 D10 747 D10 757 72S 757 S80 757 D10 M80 72S 72S D10 72S S80 D10 72S D10 72S 757 D10 S80 733 757 72S 72S M80 733 72S 733 72S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (57 of 67) [19/1/2003 1:40:23]

F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y B F Y F Y B F Y M F Y B F C Y F C Y F Y B FN YN BN F Y B F Y B FN Y B F Y B F Y B F Y B F M V F Y B F Y B F Y B F Y B F Y B F Y B F Y B A Y Q F Y B F C Y F Y B F Y B FN YN B Y B H F Y B F Y B Y B H F Y B Y B H F Y B

M M M M M M M M M M M M M M M M M M M M M M M M M

Q H Q Q H H Q H H Q H Q Q Q H H Q Q H Q Q H H Q Q

M Q M B B M M M M H M M Q K H M M H M M M H H B H M M Q M M K M K M

Q B H M M H H Q Q M H Q M M Q Q M Q Q Q K M M M Q Q K Q H Q H Q H

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

HP1235 DL1710 HP 518 UA1557 NW1238 AS 608 AA 716 UA1065 AS 30 US2164 UA1415 AS 146 UA1751 UA1149 AS 150 UA 839 UA1145 AS 188 UA1759 UA1761 UA1071 AS 88 UA1709 US2168 UA1557 AS 176 AS 196 UA1785 UA1779 US 603 AS 94 UA1405 UA1781 TW 80 TW 326 TW 150 TW 100 DL 170 WN 968 HP 215 WN 962 WN 466 WN 964 DL 152 DL 618 DL 331 DL 643 DL 134 DL 156 AA 566 AA 530 AA 566 US1118 UA 92 NW 182 NW 352 TW 754 UA 132 NW 326 AA 530 UA 20

SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO

147P 230P 400P 430P 510P 615P 940A 625A 700A 710A 730A 745A 830A 930A 940A 1030A 1130A 1145A 1230P 130P 230P 250P 330P 425P 430P 435P 520P 530P 630P 655P 659P 730P 830P 112A 644A 1004A 115P 830A 635A 810A 1020A 1205P 600P 530A 700A 830A 1205P 330P 1020P 700A 1222P 700A 705A 800A 800A 820A 825A 1000A 1125A 1222P 1235P

PHX PHX PHX PHX PHX PHX PVD SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO STL STL STL STL TPA ABQ ABQ ABQ ABQ ABQ ATL ATL ATL ATL ATL ATL BNA BNA BOS BOS BOS BOS BOS BOS BOS BOS BOS BOS

700P 756P 750P 949P 1021P 955P 741P 829A 905A 909A 927A 940A 1034A 1133A 1145A 1233P 129P 150P 230P 332P 428P 450P 522P 621P 622P 635P 725P 729P 830P 851P 900P 921P 1026P 657A 1239P 357P 715P 843P 1145A 1243P 135P 510P 915P 225P 554P 348P 736P 1047P 539A 112P 638P 519P 627P 429P 433P 612P 450P 819P 935P 1054P 855P

S S S D D D LD B B B B B B S S L L L L S S L S S D D D D D D D # # S B L L BL

B BS B L D S R L RS LD BS BS # B LD LD LD L

1 1 0 1 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 2 1 0 2 0 1 2 0 0 0 0 0 0 1 2 0 0 1 0 1 1 1 0

73S 72S 733 72S 320 72S S80 733 M80 733 72S 727 733 733 M80 735 72S 72S 73S 73S 733 72S D10 733 72S 72S M80 733 73S 733 72S 72S 73S M80 M80 M80 M80 72S 73S 73S 73S 733 73S L10 757 L10 767 757 763 S80 S80 S80 733 757 757 757 767 D10 D10 S80 D10

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (58 of 67) [19/1/2003 1:40:24]

Y F Y F FN F F F F F F F F F F F F F F F F F F F F F F F F F F F F FN F F F F Y Y Y Y Y F@ F F F F FN F F F F F F F F F F F F

B H K Q Y B M Q B H K Q Y B M Q YN BN M H Y B M H Y B M H Y B M Q Y B M H Y B H Q Y B M Q Y B M H Y B M Q Y B M Q Y B M H Y B M Q Y B M Q Y B M H Y B M Q Y B M Q Y B M Q Y B M H Y B M Q Y B H Q Y B M Q Y B M H Y B M H Y B M Q Y B M Q Y B H Q Y B M H Y B M Q Y B M Q YN B Q M Y B Q M Y B Q M Y B Q M Y B M Q K L B Q B H K Q K L B Q K L B Q K L B Q Y@ B@ M@ Q@ Y B M Q Y B M Q Y B M Q Y B M Q YN BN MN QN Y B H M Y B H M Y B H M Y B H Q Y B M Q Y B M H Y B M H C Y B Q Y B M Q Y B M H Y B H M Y B M Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

NW 356 TW 44 CO 256 AA1256 US1492 US1118 CO 454 US 626 UA 654 CO 196 CO 818 UA 970 UA 920 CO 838 UA 748 UA 222 CO1632 UA 506 UA 490 UA 582 DL 374 AA 580 AA1256 DL 618 AA 70 AA 786 DL1014 AA 963 UA 478 AA 240 DL 180 AA 254 DL 852 NW 340 NW 342 NW 344 NW 346 WN 955 WN 969 WN 957 AA 64 UA 26 DL 110 TW 842 AA 16 DL1426 UA 18 AA 20 TW 44 UA 24 AA 18 DL 808 UA1731 US1492 DL 987 PA 445 UA1101 UA1701 DL 823 US2167

SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO

230P 1000P 1155P 640A 700A 705A 1222P 800P 630A 630A 855A 902A 1118A 1200N 230P 231P 310P 420P 705P 855P 1220A 1225A 640A 700A 808A 945A 1055A 1102A 1243P 125P 230P 500P 555P 810A 1220P 140P 1100P 815A 930A 1015A 800A 800A 800A 800A 1230P 1255P 120P 330P 1000P 1010P 1015P 600A 630A 700A 700A 700A 700A 730A 800A 800A

BOS BOS BOS DCA DCA DCA DCA DCA DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DEN DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DTW DTW DTW DTW HOU HOU HOU JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK JFK LAX LAX LAX LAX LAX LAX LAX LAX LAX

1212A 754A 1000A 447P 546P 418P 1059P 809A 952A 952A 1225P 1224P 238P 318P 551P 556P 632P 743P 1029P 1216A 523A 541A 1156A 1222P 131P 301P 415P 417P 615P 653P 748P 1013P 1102P 335P 749P 904P 616A 245P 435P 540P 423P 418P 413P 420P 846P 1045P 942P 1132P 600A 622A 610A 704A 748A 810A 807A 826A 820A 850A 915A 910A

D S # # SB BL LD S B B B B L L S S S S D # S B B B L L L L L L D D B L L S

R BS B B L LD L D S S S S S S S S S S S S

1 1 1 1 2 1 1 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 2 2 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0

757 L10 733 757 733 733 M80 734 72S M80 733 72S D10 AB3 72S 757 AB3 735 733 733 757 757 757 757 767 767 767 757 733 D10 763 767 767 D10 757 72S 757 733 73S 733 767 D10 310 L10 767 757 767 767 L10 767 D10 757 72S 733 73S AB3 D10 737 73S 733

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (59 of 67) [19/1/2003 1:40:24]

F F A F F F A FN F A Y F F F F F F F F Y FN FN F F F F F F Y F F F F F F F FN Y Y Y F F F F F F F F F FN FN F F F F F F F F F

Y B M H C Y B Q Y Q H K Y B H M Y B H Q Y B H Q Y Q H K YN BN HN QN Y B M Q Y Q H K Q H K B Y B M Q Y B M Q A Y Q H Y B M Q Y B M Q A Y Q H Y B M Q Y B M Q B M Q H YN BN MN QN YN B H M Y B H M Y B M Q Y B H M Y B H M Y B M Q Y B H M B M Q H Y B H M Y B M Q Y B H M Y B M Q Y B M H Y B M H Y B M H YN BN M H K L B Q K L B Q K L B Q C Y B H Y B M Q C Y B M C Y B Q Y B H M Y B M Q Y B M Q Y B H M C Y B Q YN B M Q Y B H M Y B M Q Y B M Q Y B H Q Y B M Q Y B M Q Y B M Q Y B M Q Y B M Q Y B H Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

PA 481 UA1103 UA1703 US2249 DL 933 UA 936 UA1705 DL1025 US 206 UA1107 UA1707 DL1107 UA 844 US 268 UA 806 US 264 DL1223 UA1111 TW 223 UA 810 DL 163 US2170 UA1115 UA1715 US2121 UA1117 DL 275 UA1717 UA1119 US2393 DL 387 TW 177 UA1719 DL1400 DL 453 US 390 UA1121 UA1721 US2192 DL 569 UA1123 UA1723 US2188 DL 642 UA1125 UA1725 DL 979 US2399 UA1127 DL 805 US 626 UA1129 PA 447 DL 963 US 610 UA1131 US2389 US2397 PA 445 AA1540 TW 492

SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO

800A 800A 830A 900A 900A 900A 930A 1000A 1000A 1000A 1030A 1100A 1100A 1100A 1130A 1200N 1200N 1200N 1214P 1230P 100P 100P 100P 130P 200P 200P 200P 230P 300P 300P 300P 310P 330P 330P 400P 400P 400P 430P 500P 500P 500P 530P 600P 600P 600P 630P 700P 700P 700P 800P 800P 800P 830P 900P 900P 900P 1000P 1100P 700A 730A 1010A

LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX LAX MIA MIA MIA

924A 923A 953A 1015A 1016A 1020A 1050A 1112A 1116A 1118A 1150A 1210P 1220P 1216P 1246P 110P 111P 116P 130P 151P 215P 218P 221P 247P 310P 317P 312P 346P 416P 410P 411P 423P 449P 441P 511P 510P 519P 547P 610P 610P 617P 647P 710P 711P 717P 749P 812P 810P 819P 915P 910P 919P 956P 1006P 1010P 1017P 1116P 1210A 519P 345P 838P

S S S S S S S # S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S #

SB R LD

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1

72S 735 73S 733 73S 735 73S 73S 733 72S 73S 73S 73S 733 733 733 73S 737 M80 73S 73S 733 735 73S 733 73S 73S 73S 73S 734 73S L10 733 72S 73S 733 73S 737 733 73S 73S 735 733 73S 73S 73S 73S 733 737 73S 734 735 AB3 73S 733 72S 734 733 AB3 767 L10

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (60 of 67) [19/1/2003 1:40:24]

F F F F F F F F F F Y F Y F Y F F F F F F F F F F F F F F F F F F F F F F F F F F F F F Y F F F Y F FN F F FN FN F FN FN F F F

Y Y Y Y Y Y Y Y Y Y B Y B Y B Y Y Y Y C Y Y Y Y

B B B B B B B B B B M B M B M B B B B Y B B B B

M M M H M M M M H M Q M Q H Q H M M Q B M H M M

Q Q Q Q Q Q Q Q Q Q H Q H Q H Q Q Q M M Q Q Q Q

Y Y Y Y Y Y C Y Y Y Y Y Y Y Y Y Y Y Y B Y Y Y B Y YN Y Y YN YN Y YN YN Y C C

B B B B B B Y B B B B B B B B B B B B M B B B M B BN B B BN BN B BN BN B Y Y

M M M M H M B M M M H M M H M M M H M Q M M H Q M HN M M MN HN M HN HN M B B

Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q Q H Q Q Q H H QN Q Q QN QN Q QN QN Q M Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

UA 378 PA 447 NW 928 NW 350 NW 352 NW 326 NW 356 NW 358 DL 618 UA 372 AA 580 AA 778 UA 134 AA1148 UA 974 UA 438 AA1098 AA 42 UA 132 UA 126 UA 122 AA 220 UA 808 UA 130 AA 838 UA 194 UA 818 UA 136 UA 128 AA 484 US1776 NW 340 DL 336 US 30 UA 94 US 624 WN 456 WN 968 HP 897 HP1530 UA1285 HP 594 WN 466 WN 658 UA1243 HP1406 WN1660 HP1236 WN 791 HP1260 UA1557 WN 429 UA1575 UA1750 AS 143 UA1468 UA1250 US2106 AS 93 UA1430

SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO

145P 830P 130A 700A 820A 1125A 230P 415P 700A 130P 1225A 620A 650A 749A 805A 820A 845A 930A 1000A 1015A 1100A 1245P 110P 111P 303P 420P 630P 1121P 1122P 1140P 745A 810A 1215P 125P 150P 1140P 620A 635A 655A 930A 1100A 1153A 1205P 1230P 125P 225P 400P 520P 615P 632P 700P 830P 905P 630A 730A 730A 830A 905A 915A 930A

MIA MIA MSP MSP MSP MSP MSP MSP MSY MSY ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD PHL PHL PHL PHL PHL PHL PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX PHX SEA SEA SEA SEA SEA SEA SEA

1009P 631A 700A 1230P 144P 450P 755P 939P 250P 744P 915A 1233P 1251P 200P 215P 415P 235P 320P 356P 412P 453P 716P 715P 715P 915P 1021P 1229A 506A 516A 551A 341P 615P 1005P 1056P 943P 1002A 910A 1020A 950A 1221P 150P 244P 345P 330P 418P 516P 650P 926P 915P 925P 949P 1125P 1157P 831A 930A 933A 1035A 1110A 1115A 1136A

D S S B B L D D BS L B B B B B L R L L L L L L L D D D # # # B B L L L S

# # L S

# #

S D # B B B B S B S

0 1 0 0 0 0 0 0 1 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 0 1 0 1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0

757 AB3 757 D10 757 D10 757 757 757 733 757 S80 747 S80 757 72S 767 767 D10 D10 72S 757 747 72S S80 757 757 D10 757 S80 767 D10 757 767 D10 733 73S 73S 733 320 72S 320 733 735 72S 320 73S 733 733 733 72S 73S 733 73S M80 73S 737 733 72S 733

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (61 of 67) [19/1/2003 1:40:24]

F F F FN F F F FN F F F F F F F F F F F F Y F F F F F F FN FN F F F F F F FN Y Y F F F F Y Y F F Y Y Y F F Y F F F F F F F F

Y B M Q Y B M Q Y B M H YN BN M H Y B M H Y B M H Y B M H YN BN M H Y B M Q Y B M Q YN B M Q Y B M H Y B M Q Y B M H Y B M Q Y B M Q Y B M H Y B M H Y B M Q Y B M Q B M Q H Y B M H C Y B M Y B M Q Y B M H Y B M Q C Y B M YN B M Q YN B M H Y B M H Y B H Q Y B M H Y B M Q Y B H Q Y B M Q YN BN HN QN K L B Q K L B Q Y B H K Y B H K Y B M Q Y B H K K L B Q K L B Q Y B M Q Y B H K K L B Q B H K Q K L B Q Y B H K Y B M Q K L B Q Y B M Q Y B M Q Y B M H Y B M Q Y B M Q Y B H Q Y B M H Y B M Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AS 147 UA1552 UA1760 UA 838 UA1108 AS 87 UA1706 UA1776 US2173 AS 195 UA1770 AS 157 UA1716 UA1170 US2171 AS 31 UA1774 TW 324 TW 152 TW 492 TW 180 HP1406 DL 808 UA 974 US 268 NW 28 US 610 TW 297 WN 343 TW 123 TW 845 TW 89 DL 828 TW 646 DL 967 TW 518 TW 360 DL1430 TW 534 DL1434 DL 611 TW 572 AA1628 TW 696 TW 594 AA 687 TW 552 AA 584 TW 686 TW 196 TW 436 TW 810 AA 278 TW 634 TW 150 TW 184 TW 80 TW 410 TW 358 TW 262 TW 700

SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO SFO STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL

1025A 1030A 1130A 1230P 130P 225P 230P 330P 410P 425P 430P 530P 530P 630P 755P 825P 920P 110A 640A 1010A 120P 225P 600A 805A 1100A 1245P 900P 917A 940A 1157A 640P 924P 630A 739A 1010A 1030A 115P 115P 449P 500P 759P 812P 631A 821A 1030A 1202P 133P 535P 802P 741A 1036A 136P 220P 449P 503P 800P 749A 1035A 1149A 1233P 125P

SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA SEA STL STL STL STL STL TPA TPA TPA TPA TPA ABQ ABQ ABQ ABQ ABQ ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL BNA BNA BNA BNA BNA BNA BNA BOS BOS BOS BOS BOS BOS BOS DCA DCA DCA DCA DCA

1220P 1232P 136P 240P 336P 420P 432P 532P 610P 620P 628P 725P 732P 832P 955P 1020P 1122P 652A 1218P 349P 705P 1011P 337P 649P 824P 1025P 650A 1102A 1110A 140P 825P 1111P 858A 1012A 1235P 106P 357P 352P 726P 726P 1023P 1054P 731A 929A 1136A 105P 250P 637P 918P 1109A 210P 510P 747P 1009P 846P 1125P 1040A 132P 241P 332P 428P

L L L L S L S S S D S D D D S S # S B L L # SB BS SL L S S # D # #

S # # #

# #

B # S D D D S B # # # #

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0

727 72S 733 735 735 72S 737 73S 733 M80 72S 72S 73S 733 733 M80 733 M80 L10 L10 L10 320 757 757 733 72S 733 72S 73S M80 L10 M80 D9S D9S M80 D9S D9S 72S D9S 72S D9S D9S 727 727 72S 72S D9S 72S D9S M80 M80 L10 S80 M80 M80 L10 M80 M80 72S 72S 72S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (62 of 67) [19/1/2003 1:40:24]

F F F F F F F F F Y F F F F F F F FN F F F F F F F Y FN F Y F F FN F F F F F F F F FN F F F F F F F F F F F F F F F F F F F F

Y B M H Y B M Q Y B M Q Y B M Q Y B M Q Y B M H Y B M Q Y B M Q Y B H Q B M H Q Y B M Q Y B M H Y B M Q Y B M Q Y B H Q Y B M H Y B M Q YN B Q M C Y B Q C Y B Q C Y B Q Y B H K Y B Y B M Q Y B H Q B M H Q YN BN HN QN Y B Q M K L B Q Y B Q M C Y B Q YN B Q M Y B M Q Y B Q M Y B M Q Y B Q M Y B Q M Y B M Q Y B Q M Y B M Q YN BN MN QN Y B Q M Y B H M Y B Q M Y B Q M Y B H M Y B Q M Y B H M Y B Q M Y B Q M Y B Q M C Y B Q Y B H M Y B Q M Y B Q M C Y B Q Y B Q M Y B Q M Y B Q M Y B Q M Y B Q M

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

TW 440 TW 10 TW 280 TW 240 UA 679 TW 457 UA 165 TW 399 UA1233 TW 263 UA 339 TW 19 TW 703 AA 801 TW 531 AA 891 TW 525 AD1271 TW 551 AA 501 TW 579 AD1218 TW 621 AA 217 TW 657 AA 559 AA1405 TW 445 NW 460 TW 432 NW 462 TW 648 TW 224 NW 466 WN 356 TW 220 NW 464 TW 320 WN 531 TW 605 WN 653 WN 469 WN 345 TW 309 WN 589 WN 759 TW 343 WN 351 TW 639 WN 663 TW 405 TW 844 TW 700 TW 60 TW 634 TW 10 TW 443 TW 223 WN 343 TW 177

STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL

137P 505P 604P 812P 640A 917A 1001A 1158A 320P 324P 635P 636P 930P 655A 743A 755A 922A 1000A 1157A 105P 116P 205P 309P 350P 628P 650P 840P 921P 720A 739A 935A 1030A 130P 135P 345P 450P 450P 845P 845A 912A 1015A 1115A 1140A 1209P 130P 400P 411P 540P 624P 920P 925P 118P 125P 146P 449P 505P 902A 907A 940A 1210P

DCA DCA DCA DCA DEN DEN DEN DEN DEN DEN DEN DEN DEN DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DFW DTW DTW DTW DTW DTW DTW DTW DTW DTW DTW HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU HOU JFK JFK JFK JFK JFK LAX LAX LAX LAX

442P 800P 900P 1108P 757A 1050A 1123A 128P 448P 500P 753P 807P 1055P 844A 939A 946A 1123A 140P 152P 251P 313P 545P 501P 546P 827P 849P 1021P 1115P 948A 1005A 1205P 1259P 408P 403P 705P 731P 719P 1114P 1040A 1124A 1210P 210P 140P 231P 335P 545P 624P 740P 846P 1115P 1134P 450P 617P 520P 821P 938P 1109A 130P 1225P 423P

# D D # B S S # S S D D B B B

# # #

D D

S # S S

S S

S

#

S D

# # S D D B B L

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 1 1 1

D9S 72S M80 M80 72S 72S 72S M80 737 72S 72S 72S 72S 727 D9S S80 D9S SWM D9S S80 D9S SWM D9S S80 D9S S80 S80 DC9 DC9 72S 72S 72S M80 D9S 73S 72S D9S 72S 733 72S 735 733 73S D9S 73S 73S 72S 73S D9S 733 72S 767 72S 72S M80 72S 72S M80 73S L10

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (63 of 67) [19/1/2003 1:40:24]

F F F F F F F Y F F F F FN F F F F Y F F F Y F F F F F FN F F F F F F Y F F F Y F Y Y Y F Y Y F Y F Y FN F F F F F F F Y F

Y Y Y Y Y Y Y B Y Y Y Y YN Y Y Y Y K Y Y Y K Y Y Y Y YN YN Y Y Y Y Y Y K Y Y Y K Y K K K Y K K Y K Y K YN C Y Y Y Y Y Y K C

B B B B B B B Q B B B B B B B B B Q B B B Q B B B B B B B B B B B B L B B B L B L L L B L L B L B L B Y B B B B B B L Y

Q Q Q Q M Q M M M Q M Q Q H Q H Q V Q H Q V Q H Q H H Q M Q M Q Q M B Q M Q B Q B B B Q B B Q B Q B Q B Q Q Q Q Q Q B B

M M M M Q M Q V Q M Q M M M M M M H M M M H M M M M M M H M H M M H Q M H M Q M Q Q Q M Q Q M Q M Q M Q M M M M M M Q Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

TW 31 TW 403 TW 721 TW 171 AA1363 TW 412 TW 496 AA 687 TW 246 TW 204 TW 492 NW 940 TW 336 NW 707 TW 591 NW 592 TW 507 TW 516 TW 693 NW 594 NW 598 TW 331 NW 596 TW 473 TW 539 TW 633 TW 651 TW 385 UA 984 AA1363 TW 378 AA 324 UA 288 AA 528 TW 670 UA 840 TW 326 AA 278 UA 196 TW 234 TW 574 UA 728 AA 402 TW 88 TW 542 US 272 TW 430 UA 288 US 881 TW 344 TW 756 US 522 TW 108 TW 446 HP 851 WN 517 TW 169 WN 496 TW 59 WN 487 HP 480

STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL

1220P 327P 650P 927P 710A 848A 1030A 1202P 130P 449P 458P 550P 808P 700A 902A 1045A 1200N 121P 317P 330P 520P 639P 800P 917P 817A 1157A 307P 702P 611A 710A 803A 830A 910A 1128A 1134A 1145A 132P 220P 303P 321P 509P 551P 555P 738P 803P 615A 745A 910A 1015A 1042A 125P 305P 532P 806P 630A 800A 902A 1015A 1157A 120P 148P

LAX LAX LAX LAX MIA MIA MIA MIA MIA MIA MIA MIA MIA MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSP MSY MSY MSY MSY ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD ORD PHL PHL PHL PHL PHL PHL PHL PHL PHL PHX PHX PHX PHX PHX PHX PHX

217P 527P 854P 1129P 1248P 1231P 209P 525P 513P 1034P 838P 1055P 1247A 830A 1037A 1218P 136P 301P 457P 509P 655P 815P 931P 1049P 1009A 149P 459P 901P 715A 818A 915A 945A 1024A 1232P 1245P 1254P 248P 333P 415P 435P 626P 715P 720P 856P 921P 908A 1044A 159P 108P 151P 433P 558P 846P 1114P 845A 1105A 1123A 1230P 220P 425P 410P

L D D S # B # # D D

S

S S

S

B # D S # # # S # # # #

# # # # # # B B SL S S # S D B B L S

0 0 0 0 1 0 0 1 0 2 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 1 0

L10 L10 747 L10 72S M80 72S 72S M80 M80 L10 D9S M80 D9S D9S D9S D9S D9S D9S D9S D9S 72S D9S D9S D9S M80 D9S D9S 72S 72S 727 S80 72S S80 D9S 72S M80 S80 733 72S M80 733 S80 M80 D9S 100 72S 72S 100 72S L10 100 72S 72S 320 73S 72S 733 72S 73S 733

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (64 of 67) [19/1/2003 1:40:24]

F F F FN F F F F F F F F F FN F F F F F F F F F FN F F F F F F F F F F F F F F F F F F F F F F F F F F F F F F F Y F Y F Y Y

C Y C Y C Y CN YN Y B Y B Y B Y B K Y B C Y Y B Y B YN BN Y B Y B Y B Y B Y B Y B Y B Y B YN BN YN B Y B Y B Y B Y B Y B Y M Y B Y M Y B Y M Y B Y B Y B Y M Y B Y B Y B Y B Y M Y B Y B Y B Y B Y B Y B Y B C Y Y B Y B Y B Y B K L Y B K L Y B K L B H

B B B B M Q Q M

Q Q Q Q H M M H

Q B M Q M Q M Q Q Q M M Q M Q Q Q Q Q M H Q H M H Q M Q H M Q Q M H Q Q H Q M H Q B H Q Q H B Q B Q B K

M Q H M H M H M M M H H M H M M M M M Q B M B Q B M Q M B Q M M Q B M M Q M Q Q M Q Q M M K Q M Q M Q Q

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

WN 540 TW 75 WN 545 HP 832 TW 57 TW 161 TW 694 TW 598 TW 608 TW 29 TW 53 AA 217 TW 375 TW 23 TW 223 WN 653 TW 177 WN 487 TW 183 HP 607 TW 71 TW 67 TW 222 TW 226 NW 880 TW 456 TW 204 TW 336 DL 581 AA1111 DL 868 DL 422 DL 436 DL 578 DL 990 DL 412 DL 197 DL 552 DL1128 DL 582 AA1357 AA1475 AA 483 AA1162 US1586 US1617 DL 208 NW1298 US 718 US1296 US 442 DL 514 US 728 NW1842 US1248 US 442 NW1844 TW 457 CO 119 DL 581

STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL STL TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA

255P 312P 425P 610P 638P 926P 739A 115P 817P 917A 316P 350P 624P 918P 907A 1015A 1210P 120P 325P 402P 630P 924P 839A 1100A 1155A 125P 449P 808P 600A 717P 610A 725A 935A 1115A 1255P 220P 410P 610P 750P 930P 815A 140P 520P 725P 835A 1035A 1140A 1200N 1225P 135P 515P 710P 830A 1220P 100P 515P 520P 700A 220P 600A

PHX PHX PHX PHX PHX PHX RDU RDU RDU SEA SEA SEA SEA SEA SFO SFO SFO SFO SFO SFO SFO SFO TPA TPA TPA TPA TPA TPA ABQ ABQ ATL ATL ATL ATL ATL ATL ATL ATL ATL ATL BNA BNA BNA BNA BOS BOS BOS BOS BOS BOS BOS BOS DCA DCA DCA DCA DCA DEN DEN DFW

550P 537P 640P 837P 901P 1148P 1029A 406P 1114P 1200N 557P 858P 901P 1150P 1139A 320P 225P 540P 536P 743P 846P 1154P 1159A 221P 425P 451P 810P 1128P 905A 1043P 730A 856A 1102A 1244P 222P 342P 537P 729P 910P 1047P 907A 230P 818P 817P 1123A 309P 224P 250P 318P 556P 927P 954P 1033A 229P 258P 718P 728P 1050A 525P 733A

S S D S B # B D D D S B L D S D S B # S # D # D # # S S

S

B # D D B L L L L S S D B L L S D BS S B

1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 2 0 2 0 1 0 0 0 0 1 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 0 0 0 1 1 0 0 0 0 0 0 1 1 0

733 72S 733 733 72S 72S D9S D9S D9S M80 M80 S80 M80 M80 M80 735 L10 73S L10 73S L10 M80 72S 72S D9S D9S M80 M80 M80 72S 72S 763 M80 72S 72S 757 763 72S 72S 763 72S 72S 72S 72S M80 734 757 72S 73S 734 733 757 734 D9S 73S 733 D9S 72S 733 M80

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (65 of 67) [19/1/2003 1:40:24]

Y F Y Y F FN F F F F F F F FN F Y F Y F Y F FN F F F F F F F@ F FN F F F F F F F F FN F F F F F F F F F F F F F FN FN F F F A FN

K Y K B Y YN Y Y Y Y Y Y Y YN Y K C K C B C YN Y Y Y Y Y Y Y@ Y YN Y Y Y Y Y Y Y Y YN Y Y Y Y Y Y Y Y Y Y Y Y Y YN YN Y Y Y Y YN

L B L H B B B B B B B M B B B L Y L Y H Y B B B B B B B B@ H BN B B B B B B B B BN B B H B B B B B B B B B B BN BN B B B Q BN

B Q Q M B Q K Q Q M Q M Q M Q M Q M Q M Q M B H Q M Q M Q M B Q B Q B Q B Q K Q B Q Q M Q M Q M M H Q M Q M Q M M@ Q@ B M MN QN M Q M Q M Q M Q M

Q

M Q MN QN M H M H B M M H H Q H Q M Q M H H Q H Q H Q M Q H Q M H HN QN H Q M H Q M H K MN QN

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AA 311 AA 443 DL 149 DL1285 AA 719 AA 476 DL 219 AA1111 DL 868 NW 879 NW 479 NW 481 NW 483 DL 412 NW 487 NW 485 US1644 TW 304 DL 477 US 144 CO1093 US1810 DL 141 US1416 US1533 PA 985 US1081 US 785 AA1657 US 768 PA 499 US1199 US 738 PA 319 TW 204 US1588 TW 336 NW 447 NW 449 DL 324 NW 489 DL 477 CO 119 UA 733 NW 479 TW 234 UA 177 AA1475 AA 476 UA 769 US 524 DL 276 US1617 US 673 US 15 US 62 US 854 DL 219 AA1630 AA 702 DL 412

TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA TPA

700A 825A 920A 1255P 120P 400P 415P 717P 610A 745A 855A 1000A 115P 220P 510P 630P 110P 206P 745A 835A 300P 535P 555P 645A 840A 915A 1125A 150P 252P 330P 430P 440P 705P 800P 845P 905P 1159P 835A 425P 545P 715P 745A 220P 710A 855A 1254P 1255P 140P 400P 749P 645A 1000A 1035A 1155A 235P 605P 755P 415P 636A 100P 220P

DFW DFW DFW DFW DFW DFW DFW DFW DTW DTW DTW DTW DTW DTW DTW DTW JFK JFK LAX LAX LAX LAX LAX MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MIA MSP MSP MSP MSP MSY MSY ORD ORD ORD ORD ORD ORD ORD PHL PHL PHL PHL PHL PHL PHL PHX RDU RDU RDU

834A 1000A 1052A 228P 256P 542P 556P 846P 1010A 1237P 1134A 1249P 355P 955P 751P 906P 522P 450P 1059A 1102A 615P 802P 753P 739A 935A 1013A 1220P 245P 350P 425P 528P 535P 800P 858P 1034P 1000P 1247A 1050A 646P 932P 934P 812A 254P 855A 1235P 435P 231P 444P 915P 931P 859A 140P 1259P 209P 449P 949P 1019P 820P 816A 235P 600P

B B B L # S S D # S B L L S D D L S B B # D D # #

#

B D D D B B B # L # SD S B L L L L DS S SD B L S

0 0 0 0 0 0 0 0 1 1 0 0 0 3 0 0 1 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1 1 0 1 1 0 0 1 0 0 0 1 0 1 0 0 1

72S 72S 763 M80 72S 72S 72S 72S 72S 72S 72S 72S M80 757 D9S 72S 734 72S 763 733 733 733 757 733 F28 72S 73S F28 72S 73S 72S 73S F28 72S M80 734 M80 72S 72S M80 72S 763 733 757 72S 72S 72S 72S 72S 757 M80 72S 734 M80 M80 733 734 72S 72S 72S 757

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (66 of 67) [19/1/2003 1:40:24]

F Y B H M F Y B H M F Y B M Q F Y B M Q F Y B H M F Y B H M F Y B M Q F Y B H M F@ Y@ B@ M@ Q@ F Y B M H FN YN BN M H F Y B M H F Y B M H F Y B F Y B M H F Y B M H F Y B H Q F Y B Q M F Y B M Q Y B H Q M Y H K B V F Y B H Q Y B M Q H F Y B H Q F Y B H M F Y B M Q F Y B H Q F Y B H M F Y M B Q F Y B H Q F Y B M Q F Y B H Q Y B H Q M F Y B M Q F Y B Q M F Y B H Q FN YN B Q M F Y B M H F Y B M H F Y B M Q F Y B M H F Y B M Q A Y Q H K F Y B M Q F Y B M H F Y B Q M F Y B M Q F Y B H F Y B F Y B M Q F Y B H Q F Y B M Q F Y B H Q F Y B H Q F Y B H Q F Y B H M F Y B H Q F Y B M Q F Y B H M F Y B H M F Y B

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE

AA 483 DL1063 UA 177 US1260 DL 197 US 144 US1810 DL 141 TW 457 AA1357 TW 245 TW 234 TW 377

TPA 520P TPA 635A TPA 1255P TPA 335P TPA 410P TPA 835A TPA 535P TPA 555P TPA 700A TPA 815A TPA 945A TPA 1254P TPA 426P

RDU SEA SEA SEA SEA SFO SFO SFO STL STL STL STL STL

701P 1235P 551P 1007P 905P 116P 1016P 1017P 831A 1102A 1118A 221P 554P

D BL LD SD D BS D D B B B # D

0 2 1 2 1 1 1 1 0 1 0 0 0

72S 757 72S 733 763 733 733 757 72S 72S M80 72S 72S

file:///E|/WEBSITE/BIBLIO/TESTDATA/AIRPLANE (67 of 67) [19/1/2003 1:40:24]

F F F F F Y F Y F F F F F

Y Y Y Y Y B Y B Y Y Y Y Y

B B B B B H B M B H B B B

H M M H M Q H Q Q B Q Q Q

M Q Q Q Q M Q H M M M M M

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

---------CHINESE An, Ven F Cai, Shanghe Cai, Sui L Cai, Tian X Cao, Chongguang Cao, Jia-Ding Cao, Xi-hua Chen, Caoyu Chen, Dayue Chen, De H Chen, Gong-ning Chen, Han-lin Chen, Jiecheng Chen, Lan S Chen, Min Chen, Ming K Chen, Mu-Fa Chen, P D Chen, Peixin Chen, Rong S Chen, Shao Z Chen, Shuping Chen, Shuxing Chen, Sui Y Chen, Wende Chen, William Y Chen, Yi X Chen, Yong G Chen, Yubo Chen, Zengrong Chen, Zhi J Chen, Zhi-Min Chen, Zhihua Chen, Zhonghu Chen, Zuchi Cheng, An S Cheng, Fuchang Chengye, You Chung, Si-Kit Ding, Hua Ding, Ren Ding, Tong-Ren Ding, Wei Y Ding, Xie P Dong, Yu X Dong, Zhen X Duan, Guang-Ren Erxiong, Jiang Fang, Ai N Fang, Hongjin Fang, Jin X Fang, Nengwen Feng, Bei Y Feng, Minfu Feng, Yu Y Fu, Changlin Fu, Chu L Fu, Dong X Fu, Shusheng file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (1 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Gao, Chunyi Gao, Guo Z Gao, Shi-an Gao, Zhanhai Geng, Xianguo Gong, Guang L Gong, Sheng Gong, Weibang Guan, Ping Guo, Dajun Guo, Shun S Guo, Xiao F Guo, Xin K Hai, Ding Haibao, Duan Han, Tian X Han, Zheng Z He, Daichuan He, Sheng W He, Xiao L Hong, Jia X Hong, Yuan Hou, Chengjun Hou, Jin C Hou, Zi-Xin Hsiao, L Hu, Xiao D Hu, Zejun Hua, Xin-Hou Huang, Cheng-Gui Huang, Jian H Huang, Lin Huang, Yongnian Huang, Zhi Y Hui, Zheng R Ji, Xinhua Jian, Renyi Jiang, Boju Jiang, Guo Y Jiang, Ji-Fa Jiang, Li S Kang, Lishan Kang, Qing D Kim, Kwangsoo Laiyi, Zhu Le, Mao Hua Lei, YingGuo Li, An-Min Li, Bingxi Li, Boyu Li, Fang Li, Feng W Li, Guiban Li, Hai Z Li, Hong X Li, Ji B Li, Jin Li, Kezheng Li, Li Li, Mengru

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (2 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Li, Qiao Li, Shaokuan Li, Shoufu Li, Shujie Li, Wantong Li, Wen R Li, XiaoMei Li, Xingmin Li, Xue L Li, Zhong Li-Shan, Liu Liang, Hua Liang, Jin Liang, Xi T Liao, Gong F Lin, Jin K Lin, Junmin Lin, Xi Lin, Yiping Lin, Yixun Lin, Yizhong Lin, Zheng-Yan Ma, Jiang H Ma, Li Ma, Tian Ma, Zhi-Ming Ma, Zhong-Qi Mao, Jing Z Nie, Yi-Yong Ouyang, Caiheng Pan, Hung-Ming Pan, Shizhong Peng, Liangang Peng, Lizhong Qi, Min Y Qian, Min Qian, Min P Qiao, Li Qin, Meng Z Qin, Tie H Qin, Zeng-fu Qingping, Cheng Qiu, Jing-Hui Qiu, Sen Ren, Ding Ren, Fu-Yao Ren, Jian H Tang, Gaohua Tang, Jian-er Teng, Zhen-huan Tian, Jinghuang Tong, Dao-Rong Wan, Zhe-Xian Wang, Bing H Wang, Bo-Ying Wang, Cheng-Shu Wang, Cun Z Wang, Daoliu Wang, De R Wang, Duo Wang, Guorong file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (3 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Wang, Hong Wang, Jia-gang Wang, Jian-pan Wang, Lian Wang, Long Wang, Mingyi Wang, Pengtao Wang, Pin C Wang, Shangzhi Wang, Sheng W Wang, Shi-kun Wang, Shi-Tau Wang, Shicheng Wang, Si-Lei Wang, Song-Gui Wang, Ting F Wang, Wei Wang, Xue-Kuan Wang, Yan M Wang, Yuanming Wang, Yuanshi Wang, Zhen-Peng Wei, Jun J Wei, Musheng Wu, Bao Qiang Wu, Cheng X Wu, Cong Wu, Jingbo Wu, Shenjian Wu, Zhuo Q Wu, Zong M Wujie, Shi Xia, Ning-Mao Xia, Zun Q Xiang, Li Xiao, Er J Xiao, Shutie Xiao, Ti J Xie, Zheng H Xin, Y L Xiong, Zhen X Xu, Daoyi Xu, Fei Xu, Jinghua Xu, Ming Y Xu, Xiaoquan Xu, Yichao Xu, Yuan-Tong Xu, Zeng K Xuan, Ti Z Xue, Weimin Yan, Jia-An Yan, Jing-hai Yan, Jurang Yan, Shi-Jian Yan, Zikun Yan, Ziqian Yang, Jin-Gen Yang, Lo Yang, Mingzhu

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (4 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Yang, Run S Yang, Shixin Yang, Shu L Yang, Weiqi Yang, Weisheng Yang, Ying C Yang, Zhong H Yanlin, Yu Yao, Bi Y Yao, Zong Y Ye, Jia-Chen Ye, Mao D Ye, Qi-Xiao Yi, Hong-Xun Yin, Jingxue Yin, Weiping Ying, Jiangang Ying, Long A You, Hong Yu, Jia-Rong Yu, Wenci Yu, Wenhuan Yu, Xijun Yuan, Ya-Xiang Yun, Tian Q Zeng, Guangxing Zeng, Yun B Zhang, Binggen Zhang, Dian Z Zhang, Fang Zhang, Fu J Zhang, Gong Q Zhang, Guan Q Zhang, Guang-Lu Zhang, Ji P Zhang, Jin H Zhang, Ke M Zhang, Shi Q Zhang, Shi S Zhang, Shouchuan Zhang, Shu-Guo Zhang, Shunian Zhang, Wei N Zhang, Weijiang Zhang, Xianke Zhang, Xue S Zhang, Zhao Z Zhang, Zhenxiang Zhang, Zhenyue Zhang, Zhi R Zhang, Zhongfu Zhangjian, Hu Zhao, Jin X Zhao, Shen Q Zheng, Dao S Zheng, Sining Zheng, Songmu Zhong, Tongde Zhou, Chaoshun Zhou, Hong X Zhou, Mei K file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (5 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Zhou, Zuo L Zhu, Jun Zhu, Yao C Zhuang, Wan Zou, Cheng Zu ---------JAPANESE Aihara, Shin I Aikawa, Hiroaki Akiyama, Kazuo Ando, Kenichi Ando, Shiro Ando, Tsuyoshi Ano, Katsunori Aoyagi, Nobuo Arai, Hitoshi Arai, Masaharu Araki, Huzihiro Arima, Satoshi Arisaka, Nakaaki Asano, Kouhei Asano, Shigemoto Asano, Takao Asano, Tetsuo Asashiba, Hideto Ashino, Ryuichi Asoo, Yasuhiro Atsuji, Masahiko Aye, Pyi Baba, Kiyoshi Bannai, Eiichi Byongmun, Kim Campbell, Mark T Chen, Yun-Gang Chikuse, Yasuko Cho, Koji Deguchi, Tetsuo Doi, Yukio Dokeh, Kotaro Doku, Isamu Doszpoly Sj, Ivan Enomoto, Kazuyuki Fasol, Gerhard Foong, See K Frankl, Peter Fuji-Hara, Ryoh Fujii, Nobuhiko Fujimagari, Tetsuo Fujioka, Hiroko Fujisaki, Rieko Fujishige, Satoru Fujita, Masayuki Fujiwara, Daisuke Fujiwara, Hidenori Fujiwara, Koji Fujiwara, Masahiko Fukuda, Koji Fukushima, Masatoshi Fukushima, Mitsuo Fukuta, Jiro

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (6 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Funabashi, Shoichi Furuta, Katsuhisa Furuta, Takayuki Furutani, Kenro Giga, Yoshikazu Goto, Hideo Goto, Kazuo Goto, Midori S Goto, Shiro Grotowski, Joseph F Hamada, Noboru Hamana, Masamichi Hamaya, Yoshihiro Hara, Shinji Hara, Takashi Harada, Shigeharu Haruki, Shigeru Hasegawa, Hidehiko Hasegawa, Keizo Hasegawa, Takayuki Hasegawa, Takemitsu Hashimoto, Kazuo A Hashimoto, Ki-ichiro Hashimoto, Takashi Hasumi, Morisuke Hatada, Kazuyuki Hatori, Osamu Hattori, Akio Hattori, Yasunao Hayami, Ken Hayashi, Kazumichi Hayashi, Mikihiro Hayashi, Nakao Hayashi, Takao Hayashi, Yoshio Hayashida, Tsuyoshi Hiai, Fumio Hicks, Craig P Hida, Takeyuki Higuchi, Yasunari Hinohara, Yukitoshi Hirabayashi, Mikihito Hirano, Norimichi Hirano, Yasuyuki Hirashita, Yukio Hirose, Hideo Hitotsuyanagi, Nobuo Fuji-Hara, Ryoh Hamada, Noboru Hamana, Masamichi Hamaya, Yoshihiro Hara, Shinji Hara, Takashi Harada, Shigeharu Haruki, Shigeru Hasegawa, Hidehiko Hasegawa, Keizo Hasegawa, Takayuki Hasegawa, Takemitsu Hashimoto, Kazuo A Hashimoto, Ki-ichiro file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (7 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Hashimoto, Takashi Hasumi, Morisuke Hatada, Kazuyuki Hatori, Osamu Hattori, Akio Hattori, Yasunao Hayami, Ken Hayashi, Kazumichi Hayashi, Mikihiro Hayashi, Nakao Hayashi, Takao Hayashi, Yoshio Hayashida, Tsuyoshi Hiai, Fumio Jimbo, Shuichi Kadota, Noriya Kaino, Keimei Kajiwara, Joji Kakiichi, Yoshiaki Kalbag, Anuroopa Kamada, Seiichi Kambayashi, Tac Kameda, Masumi Kamejima, Kohji Kamishima, Yoshinobu Kamiya, Shigeyasu Kan-on, Yukio Kanai, Masahiko Kanda, Mamoru Kaneda, Masaharu Kaneko, Akira Kaneko, Jyoichi Kaneko, Masanobu Kaneko, Tetsuo Kanemitsu, Shigeru Kanenobu, Taizo Kaneyuki, Soji Kanjin, Yuichi Kanno, Tsuneo Kano, Takeshi Karamatsu, Yoshikazu Kashihara, Kenji Kashiwagi, Yoshimi Kato, Akio Kato, Gosei Kato, Hisao Kato, Masakimi Kato, Shin-ichi Kato, Takao Katsuda, Atsushi Katsura, Toshiyuki Kawachi, Tsuneyoshi Kawahigashi, Yasuyuki Kawai, Takahiro Kawamoto, Naoki Kawamoto, Shunji Kawamura, Kazuhiro Kawasaki, Tetsuro Kawauchi, Akio Kida, Teruhiko

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (8 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Kido, Kazuo Kigami, Jun Kiho, Y Kijima, Masaaki Kikuchi, Kazunori Kikuchi, Koji Kim, Tong Ho Kimura, Shun-ichi Kimura, Tatsuo Kinukawa, Masakiti Kishi, Masanori Kita, Hiroo Kitada, Hitoshi Kitada, Yasuhiko Kiyohara, Mineo Kiyooka, Kuninori Lay, Steven R Li, Lei Maeda, Fumi-Yuki Maeda, Hironobu Maeda, Michie Maeda, Yoshiaki Maehara, Kazuhisa Maekawa, Taichi Maitani, Fumio Makino, Tetu Marumoto, Yoshihiko Maruo, Osamu Maruyama, Toru Masuda, Kyuya Masumoto, Makoto Masutomi, Fumio Matsuda, Shigeo Matsui, Kiyoshi Matsui, Shouichi Matsumoto, Yukihiro Matsumoto, Yukio Matsunobu, Seiro Matsuo, Takami Matsuoka, Yasushi Matsushita, Yasuo Matsuyama, Hiroshi Matsuyama, Yoshio Matsuzawa, Jun-ichi Matsuzawa, Tadato Matumoto, Hisayosi Matumoto, Takao Nagahara, Takasi Nagai, Osamu Nagano, Tadashi Naganuma, Daisuke Nagao, Hirosi Nagasaka, Kenji Nagase, Michihiro Nagase, Noriaki Nagata, Jun-Iti Nagata, Masatsugu Nagayama, Misao Naito, Koichiro Naitoh, Hiroo Nakada, Yutaka file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (9 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Nakagawa, Kiyokazu Nakagawa, Yoko Nakai, Mitsuru Nakajima, Haruhisa Nakajima, Kengo Nakajima, Norihiro Nakajima, Shoichi Nakajo, Yuichi Nakaki, Tatsuyuki Nakamitsu, Kuniaki Nakamura, Hiroaki Nakamura, Masahiro Nakamura, Masataka Nakamura, Riichiro Nakamura, Tokushi Nakamura, Yoshimasa Nakamura, Yoshio Nakanishi, Yasutaka Nakao, Zensho Nakashima, Masaharu Nakata, Masaomi Nakata, Mie Nakatsuka, Harunori Nakauchi, Hiroshi Nakauchi, Nobumitsu Nakayama, Akira Nakazi, Takahiko Namba, Kanji Namba, Makoto Namba, Toshiyuki Namikawa, Yukihiko Nanbu, Tokumori Naoe, Hidenori Narita, Kiyomasa Naruse, Hiroshi Narushima, Hiroshi Nehaniv, Chrystopher L Nemoto, Hiroaki Ninomiya, Hirokazu Nishida, Akira Nishida, Takaaki Nishikawa, Seiki Nishimura, Hirokazu Nishimura, Yasuichiro Nishiura, Yasumasa Nishiyama, Kyo Nodera, Takashi Noguchi, Hiroshi Noguchi, Junjiro Noguchi, Mitsunori Noguchi, Rutsu Nogura, Tsugunori Noiri, Takashi Nono, Takayuki Noriaki, Kamiya Notestine, Ronald D Kan-on, Yukio Obata, Morio Ochiai, Mitsuyuki Oda, Tadao

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (10 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Odani, Kenzi Ogawa, Hidemitsu Ogawa, Sotaro Ogawa, Toshiyuki Ogiue, Koichi Oguchi, Kunio Ogura, Yukio Ohara, Atsumi Ohguchi, Takeshi Ohno, Shuichi Ohshika, Kenichi Ohsumi, Akira Ohta, Haruto Ohta, Yoshito Ohtagaki, Hirokazu Ohtsuka, Makoto Ohwaki, Shin-Ichi Oka, Mutsuo Oka, Shinpei Okada, Ikutaro Okada, Masami Okada, Satio Okada, Soichi Okai, Takayuki Okamoto, Hisashi Okamoto, Kiyosato Okamoto, Toshiaki Okayasu, Takateru Okazaki, Ryotaro Okumura, Haruhiko Okumura, Susumu Okuyama, Akihiro Omatu, Sigeru Omori, Hideki Ono, Yoshie Onose, Hiroshi Oodaira, Hiroshi Osada, Naoki Osawa, Shingo Oshima, Kunio Ota, Schoichi Otani, Yoshihiko Otsu, Yukio Otsuka, Naohisa Otsuki, Nobukazu Owa, Shigeyoshi Oyabu, Takashi Ozawa, Kazufumi Ozeki, Michio Rao, Sasipalli V Reisewitz, Tammo M Sagara, Nobuko Saigo, Megumi Saigusa, Youichi Saito, Akira Saito, Kazuyuki Saito, Kichi-Suke Saito, Kimiaki Saito, Shiroshi Saito, Yasuyuki Saitoh, Saburou file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (11 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Sakaguchi, Shigeru Sakai, Katsuro Sakai, Kazuhiro Sakai, Makoto Sakai, Shoichiro Sakai, Takashi Sakamoto, Kunimochi Sakamoto, Kunio Sakuma, Motoyoshi Sano, Shigeru Sano, Takashi Sasaki, Hiroki Sasaki, Katsumi Sasaki, Takeshi Sasano, Kazuhiro Satake, Ichiro Sato, Hideo Sato, Hiroki Sato, Hiroshi Sato, Ken-iti Sato, Shizuka Sato, Shuichi Satoh, Takakazu Sawada, Hideki Sawami, Hideo Tachikawa, Atsushi Tada, Toshimasa Tagawa, Masa-Yoshi Taira, Kazuaki Takagi, Izumi Takahashi, Chihiro Takahashi, Joji Takahashi, Masako Takahashi, Reiji Takahashi, Shuichi Takahashi, Tetsuya Takahashi, Wataru Takahashi, Yuji Takano, Kyoichi Takayama, Nobuki Takeda, Ziro Takemoto, Hideo Takenaka, Shigeo Takeuchi, Hiroshi Takeuchi, Kisao Takeuchi, Masaru Takizawa, Kiyoshi Tamaki, Dai Tanabe, Hiroki Tanaka, Hisao Tanaka, Jun-Ichi Tanaka, Kazuyuki Tanaka, Minoru Tanaka, Shohei Tanaka, Taiyo Tanaka, Tamaki Tanaka, Yasuhiko Tanaka, Yoshihiro Tanamachi, Yoshihiro Tanigchi, Masaharu

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (12 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Tanikawa, Takao Tanno, Shukichi Tao, Yoko Tashiro, Yoshihiro Tazawa, Shinsei Teramoto, Yoshiaki Terasawa, Jun Titani, Satoko Tjoa, Iauw-Bhieng F Toda, Nobushige Tomari, Masataka Tomita, Yoshihito Tomiyama, Jun Tone, Kaoru Toyoizumi, Masao Uchino, Atsuko Uchiyama, Akihito Uchiyama, Saburo Ue, Masaaki Ueno, Kazushige Ueno, Kenji Ueno, Shuichi Ueno, Yoshiaki Ukegawa, Takasaburo Umeda, Tomio Umegaki, Hisaharu Umetani, Shin-ichi Umezawa, Toshio Unai, Yasushi U Uno, Katsuhiro Wada, Hidekazu Wada, Junzo Wada, Masaaki Wada, Toshimasa Wajima, Masayuki Wakabayashi, Nobuo Wakae, Masami Wakana, Kaoru Wakayama, Masato Wakimoto, Minoru Watabe, Mutsuo Watanabe, Hisao Watanabe, Masaji Watanabe, Nobuya Watanabe, Shoji Watanabe, Tadashi Watanabe, Tetsuro Watanabe, Toshihiro Watatani, Yasuo Yabe, Hiroshi Yagasaki, Kazuyuki Yagi, Shin-Ichi Yajima, Kenji Yamada, Hirofumi Yamada, Hiromichi Yamada, Katsuhiro Yamada, Kotaro Yamada, Miyuki Yamada, Naoki Yamada, Osanobu Yamada, Shinichi file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (13 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Yamada, Takeo Yamada, Toshihiko Yamada, Yoshio Yamagata, Shuji Yamagishi, Kikumichi Yamaguchi, Hiroshi Yamaguchi, Itaru Yamaguchi, Keizo Yamaguchi, Kohhei Yamaguchi, Seiichi Yamaguchi, Takao Yamaguchi, Yasushi Yamaguti, Kiyosi Yamaji, Minatsu Yamaki, Hiroyoshi Yamamoto, Masahiro Yamamoto, Tetsuro Yamamoto, Yoshiharu Yamano, Gosuke Yamanoshita, Tsuneyo Yamasaki, Masayuki Yamashita, Hiroshi Yamashita, Michinori Yamauchi, Manami Yamawaki, Noriaki Yamazaki, Masao Yanagawa, Minoru Yanaghiya, Akira Yanagi, Kenjiro Yanai, Hiromichi Yasuda, Toshihiko Yasue, Kunio Yasugi, Mariko Yenne, James W Yoda, Kiyoshi Yokoi, Hideo Yokonuma, Takeo Yokota, Hisashi Yokoyama, Etsuro Yokura, Shoji Yoneda, Kaoru Yoshida, Masaaki Yoshihara, Ken-Ichi Yoshikawa, Atsushi Yoshikawa, Katsuyuki Yoshimoto, Takeshi Yoshinaga, Takao Yoshino, Ken-ichi Yoshino, Masafumi Yoshino, Takashi Yoshioka, Akira Yoshise, Akiko Yoshizawa, Taro ---------KOREAN Ahn, Byung M Ahn, Inkyung Ahn, Jae M Ahn, Jung H Ahn, Sun Shin

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (14 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Ahn, Sung H Bae, Jong S Bae, Soon-Sook Bae, Sunghan Bae, Yongju Bahk, S K Baik, Young G Bang, Hyunsoo Bang, Keumseong Bang, Seung-Jin Bok, Lee H Byun, Chang Ho Byun, Yanghyun Cha, Hyung K Chae, Dongho Chang, Joo S Chang, Kun S Chang, Kun S Chang, Shin Chang, Sung K Chang, Yu-Sung Cheon, Gi-Sang Cheul, Lim H Chi, Dong P Cho, Chong-Man Cho, Gyeong-Mi Cho, Han-Hyuk Cho, Hwan G Cho, In-Ho Cho, Jung R Cho, Kwon Cho, Min-Hyung Cho, Myung Hyun Cho, Nak Eun Cho, Nam Z Cho, Nhansook Cho, Sang-Hyun Cho, Sung J Cho, Tae-Geun Cho, Yeol J Cho, Yong S Cho, Yong-Kum Cho, Young Hyun Choa, Jun S Choe, Boo R Choe, Geon H Choe, Jaigyoung Choe, Young H Choi, Bong D Choi, Byoung K Choi, Hyeong I Choi, Jeong Y Choi, Kyu H Choi, Q-Heung Choi, Sang-Il Choi, Seungil Choi, Suhyoung Choi, Sung K Choi, Sungwoo Choi, U J Choi, Un H file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (15 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Choi, Younggi Choi, Youngsoo Choi, Yuncherl Choie, Youngju Chu, Chinku Chun, Joohwan Chun, Sook-Hee Chung, Dong M Chung, Kyung T Chung, Sang K Chung, Seiyoung Chung, Soon-Yeong Chung, Taehun Chung, Youngbok Eun, Gwang S Goh, Seung-Cheol Gu, Cho I Ha, Ki S Ha, Seung Y Ha, Soon-Dock Ha, Young-Hwa Hahn, Sang-Geun Han, Chi-Geun Han, Chong-Kyu Han, Chul-Soon Han, Hyuk Han, Juncheol Her, Min Hong, Bum I Hong, Chan Y Hong, Dug Hun Hong, Seung Pyo Hong, Sung S Hong, Sungbok Hong, Bum I Hong, Chan Y Hong, Dug Hun Hong, Seung Pyo Hong, Sung S Hong, Sungbok Hong, Sungpyo Hong, Young H Im, Bokhee Im, Geun B Jang, Sun Y Jeon, Tae I Jeon, Youngmok Jeong, Ja A Jeong, Moonja Jeong, Myung-Hwa Jin, Gyo Taek Jin, Lee J Jo, Young S Ju, Hyeong-Kwan Jun, Young B Jung, Hyung C Jung, I H Jung, Il B Jyoo, Yeong-Heum Kang, Byung G

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (16 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Kang, Eunju Kang, Hyeonbae Kang, Jeongheung Kang, Joo H Kang, Nam-Gyu Kang, Pyung-Lyun Kang, Seong J Kang, Shin W Kang, Sungkwon Kang, Wan Keem, Changho Keum, Jong Hae Ki, U-Hang Kim, Agnes B Kim, Ann C Kim, Byung C Kim, Chang H Kim, Chul Kim, Dae S Kim, Dae Sig Kim, Dai-Gyoung Kim, Do S Kim, Dohan Kim, Dong-Soo Kim, Donggyun Kim, Dongsu Kim, Eden P Kim, Eung T Kim, Goansu Kim, Hae R Kim, Hee S Kim, Hong G Kim, Hong J Kim, Hong O Kim, Hong-Chul Kim, Hongchul Kim, Hoonjoo Kim, Hyang Sook Kim, Hyong-Jin Kim, Hyoung J Kim, Hyoungsoon Kim, Hyuk Kim, Hyun-Jung Kim, Hyung K Kim, Ihn Sue Kim, Jae-Gyeom Kim, Jehpill Kim, Jeong G Kim, Jeongook Kim, Jin H Kim, Jin-Soo Kim, Jintae Kim, Jong K Kim, Jong-Chul Kim, Jonglak Kim, Jongsu Kim, Keehwan Kim, Kwang I Kim, Kyung H Kim, Kyunghee Kim, Mee-Kyoung file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (17 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Kim, Myoung-Nyoun Kim, Myung-Hwan Kim, Pan S Kim, Sang M Kim, Sang Dong Kim, Sang-Bae Kim, Seok-Woo Kim, Seokchan Kim, Seon-Bu Kim, Seong-A Kim, Seung I Kim, Seunghwan Kim, Seyong Kim, Sunah Kim, Sung S Kim, Sung-Ock Kim, SungSoo Kim, Tae-Gun Kim, Tae-Hwa Kim, Wan S Kim, Wansoon Kim, Won K Kim, Yangkon Kim, Yong I Kim, Yong S Kim, Yonggu Kim, Young H Kim, Young W Kim, Young-Key Kim, Young-Kuk Kimn, H-J Lau, Jeung-Hark Lee, Chang H Lee, Chang K Lee, Chang Ock Lee, Chung N Lee, Daeshik Lee, Do H Lee, Dong W Lee, Dong-Soo Lee, Dongwoo Lee, Doobum Lee, Gyou-Bong Lee, Hei-Sook Lee, Hyun Y Lee, In-Sok Lee, Insoo Lee, Jeh G Lee, Jinoo Lee, Jong C Lee, Jong P Lee, Jong Y Lee, Jong-Bum Lee, Jong-Hyeon Lee, Joo Sung Lee, Joonyoung Lee, Jun Y Lee, Jung-Jo Lee, Jungseob Lee, Ke-Seung

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (18 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Lee, Kee Y Lee, Kwang-Bok Lee, Man H Lee, Mankeun Lee, Nam S Lee, Nyoungi Lee, Sa-Ge Lee, Sang-Gu Lee, Sang-Moon Lee, Seung-Hwan Lee, So-young Lee, Sung Chul Lee, Sungho Lee, Sungjin Lee, Sungyun Lee, Tae-Keug Lee, Woo Y Lee, Wook J Lee, Yang H Lee, Yong-ha Lee, Yongsoo Lee, Yongsoo Lee, Young W Lim, Jong In Lin, Hyun-Chon Ma, In-Sook Ma, Sangback Nam, Kwanghee Noh, Sunsook Oh, Byeong K Oh, Chang H Oh, Hi-Jun Oh, Phillial Oh, Sei-Qwon Oh, Seung J Oh, Seyoung Ohm, MiRay Pahk, Dae H Pak, Jin Suk Park, Bae H Park, Chan B Park, Chan-Young Park, Chin-Hong Park, Dae Y Park, Eunsoon Park, Hong-Suh Park, Hwasin Park, Hyungbin Park, Jae K Park, Jeong H Park, Jong A Park, Jong Y Park, Joong S Park, Joonsang Park, Ju-Hyun Park, Jun S Park, Kwang S Park, Kyewon Park, Kyoo-Hong Park, Myungkark Park, Nam S file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (19 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Park, Pil S Park, Sang-Ho Park, Sehie Park, Seungkyung Park, Sung H Park, Sung H Park, Sung J Park, SungKi Park, Sungsoo Park, Taehoon Park, Yong M Park, Young H Park, Young S Park, Young-hoon Rhee, Jung S Rhie, Gil S Ung Gi, Min Wee, In-Suk Won, Dae Y Woo, Moo H Woo, Sung-Sik Woon, Oh J Yang, Chun-Woo Yang, Hyunsuk Yang, Jae-Hyun Yang, Seung K Yim, Jin-Whan Yoo, Hyeong S Yoo, Ki-Jo Yoon, Byung J Yoon, Joung-Hahn Yoon, Ju H Yoon, Yeon S Youn, Sung-Kie Yu, Dong W Yu, Mija Yum, Bong-Jin Yum, Sangsup Yun, Gabjin Yun, Jae H ---------ITALIAN Abatangelo, L M Abate, Marco Abram, Mario R Accascina, Giuseppe Acciaro, Vincenzo Acunzo, Andrew Albano, Alberto Alessandrini, Giovanni Alliney, Stefano Altomare, Francesco Ambrosetti, Antonio Amodio, Pierluigi Ancona, Fabio Andreatta, Giovanni Anile, Angelo M Antoniazzi, Stefano Arena, Orazio Arina, Renzo

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (20 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Arioli, Gianni Arioli, Mario Arri, Paolo Arzarello, Ferdinando Aversa, Vincenzo L Avesani, Renzo Bacciotti, Andrea Baldi, Paolo Barbieri, Francesco Bardaro, Carlo Barlotti, Adriano Barozzi, Elisabetta Basile, Achille Bauer, Ingrid C Beghi, Luigi Bellen, Alfredo Beretta, Edoardo Bertero, Mario Berti, Patrizia Betori, Walter Bianchi, Gabriele Biroli, Marco Blasi, Alessandro Boffi, Giandomenico Bonomi, Ernesto Boratynski, M Boretti, Alberto A Bottazzini, Umberto Bove, Antonio Branciari, Alberto Breitung, Karl W Brenti, Francesco Brieger, Leesa M Brienza, Antonio Brugnano, Luigi Buoncristiano, Sandro Buzano, Ernesto Caccianotti, Luciano Caddeo, Renzo I Calogero, Francesco Campanino, Massimo Campiti, Michele Cannarsa, Piermarco Cantoni, Vittorio Capasso, Vincenzo Capelli, Giovanni Capelo, Antonio-Candido Capozzi, Alberto Capparelli, Stefano Capria, Carlo M Capriz, Gianfranco Caranti, A E Carfora, Mauro Caricato, Gaetano Carini, Luisa Carlucci, Maurizio Casati, Paolo Castagnola, Ercole Castagnoli, Erio A Catanese, Fabrizio M Cavallini, Fabio file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (21 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Cavicchioli, Alberto Cazzaniga, Franco Cerruti, Umberto Chareyron, Jacques Chiarenza, Filippo Chiarolla, Maria Chicco, Maurizio Cirina, Marco Coen, Salvatore Colaneri, Patrizio Collino, Alberto Conforti, Domenico Conte, Giuseppe Contessa, Maria Conti, Roberto Contro, Filippo Corsi Tani, Gabriella Coti Zelati, Vittorio Cotta-Ramusino, Paolo F Crosta, Giovanni F D'Alessandro, Paolo D'Ambra, Pasqua D'Alessandro, Paolo D'Ambra, Pasqua D'Amore, Luisa D'Aprile, Margherita D'Aquino, Paola D'Auria, Catello Dagnino, Catterina Damiani, Ernesto de Bartolomeis, Paolo De Leone, Renato De Luca, Aldo de Lucia, Paolo De Mari, Filippo De Pascale, Espedito De Sena, Silvestro F de-Michele, Leonede DeAngelis, Pasquale L Del Fra, Alberto Dell'Antonio, Gianfausto Detofoli, Rino Di Maio, Giuseppe Di Martino, Lino Di Pillo, Gianni Di Serafino, Daniela Dominici, Paolo Lanza de Cristoforis, M Van Der Mee, Cornelis V Ebagezio, Valter Elia, Michele Emmanuele, Giovanni Emmer, Michele Evangelista, Elio Del Fra, Alberto Facchinei, Francisco Facchini, Alberto Faina, Giorgio Falcone, Maurizio Famularo, Domenico

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (22 of 64) [19/1/2003 1:40:36]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Fania, Maria L Farranti, Matteo Fasano, Antonio Ferragina, Paolo Ferrari, Carlo Ferrari, Luca Ferri, Massimo Ferro, Ruggero Figa-Talamanca, Alessandro Figari, Rodolfo Filippone, Salvatore Filipponi, Piero Floreanini, Roberto Fontana, Marco Forni, Giovanni Francaviglia, Mauro Franchi, Bruno Frittelli, Marco Frosali, Giovanni Frosini, Patrizio Moseneder-Frajria, Pierluigi Paveri-Fontana, Stefano L Gaivoronski, Alexei A Galbiati, Margherita Gargano, Luisa Garrisi, Daniele Gastaldi, Tommaso Gatteschi, Luigi Gaudioso, Manlio Gaviano, Marco Gentile, Antonio Gentili, Graziano Geronazzo, Lucio Ghinelli-Smit, Dina Giambruno, Antonino Giaquinta, Mariano Gimigliano, Alessandro Giorgi, Giorgio Girone, Giovanni Giuli, Eraldo Giunta, Giulio Glielmo, Luigi Van Geemen, Bert Iannelli, Mimmo Invernizzi, Sergio Iozzi, Fabrizio Kisel, Ema De Leone, Renato Laccetti, Giuliano Laeng, Enrico Lammens, Johan M Landucci, Mario Lanteri, Antonio Lanza de Cristoforis, M Laurence, Peter M Lauria, Francesco E Leoni, Giovanni Licalzi, Marco De Mari, Filippo Di Maio, Giuseppe Di Martino, Lino file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (23 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Maffioli, Francesco Magenes, Enrico Magri, Franco Maj, Mercede Majer, Pietro E Malvenuto, Claudia Malvestuto, Francesco M Manca, Vincenzo Mangiarotti, Luigi Marchetti, Federico Marchiafava, Stefano Marcja, Annalisa Marcone, Alberto G Marconi, Carla Marino, Marina Maroscia, Paolo Martucci, Giovanni Mauceri, G C Mazzia, Francesca Mazzocca, Francesco Nacinovich, Mauro Naldi, Giovanni Naldi, Maurizio Nanni, Umberto Nannicini, Antonella Napoletani, Domenico Nardelli, Enrico Navarra, Antonio Nistri, Paolo Notarnicola, Filippo Nucci, Maria-Clara Nugari, Rita Oberto, Paolo M Oliverio, Paolo A Olla, Stefano Ornaghi, Giuseppe Orsenigo, Marta De Pascale, Espedito Di Pillo, Gianni Pacati, Claudio Pacella, Filomena Paclet, Philippe Palagi, Laura Pallara, Diego M Pandolfi, Luciano Paoluzzi, Fabrizio Pappalardi, Francesco Parmeggiani, Alberto Pasini, Antonio Pasquali, Aldo Pata, Vittorino Patrizio, Giorgio Pavani, Raffaella Pavarino, Luca Paveri-Fontana, Stefano L Peccati, Lorenzo Peloso, Marco M Pensavalle, Carlo A Pepperdine, Andrew H Perelli, Alberto

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (24 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Perrone, Antonio L Petronio, Carlo Piacentini Cattaneo, Giulia M Piazzi, Aurelio Picardello, Massimo A Picca, Domenico Piccarolo, Stefano Piccinini, Renzo A Piccinni, Paolo Piciulin, Antonio Pierantoni, Margherita Pirone, Florindo Quarteroni, Alfio M Cotta-Ramusino, Paolo F Racca, Ezio Ragaini, Enrico Rao, Salvatore Rapisarda, Paolo Regazzini, Eugenio Ricci, Fulvio Rigo, Pietro Rivara, Luigi M Rodino, Luigi Sabatini, Marco Sabatino, Pietro Sacerdote, Laura Saliani, Sandra Salinelli, Ernesto Salvadori, Anna Salvadori, Gianfausto Salvi, Zagaglia N Santarelli, Ulderico Santi, Elisabetta Sartoretto, Flavio Sbordone, Carlo Scafati Tallini, Maria Scarpi, Giambattista Scarsini, Marco Schaerf, M Scheinine, Alan Schiano, Pasquale Schoof, Rene Scolozzi, Donato Scoppola, Carlo M Scotti, Antonio Seatzu, Sebastiano Siena, Carlo Corsi Tani, Gabriella Figa-Talamanca, Alessandro Scafati Tallini, Maria Talenti, Giorgio G Tamanini, Italo Tanasi, Corrado Tancredi, Alessandro Tarantello, Gabriella Tardella, Fabio Teolis, Antonio G Tesei, Alberto Testa, Giuliano Tironi, Gino Toffalori, Carlo file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (25 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Togliatti, E G Tonet, Davide Tortorici, Adele Tozzi, Anna Travaglini, Giancarlo Trigiante, Donato Trimarco, Carmine Troi, Giancarlo Ursini, Aldo Valla, Giuseppe Van Der Mee, Cornelis V Van Geemen, Bert Ventriglia, Francesco Verhoeff, Eric F Vesentini, Edoardo Vianello, Marco Vigna, Sebastiano Vincenzi, Antonio Vinti, Calogero Viola, Carlo Volcic, A Coti Zelati, Vittorio Zacher, Giovanni G Zanco, Clemente A Zanolin, Fabio Zavatta, Giuseppe Zirilli, Francesco Zsido, Laszlo ---------HISPANIC Abellanas, Manuel Abellanas, Pedro Abia, Luis Aguirre, Julian Ahlburg, Hayo Alabert, Aureli Aldaz, Jesus M Alegre, Ignacio D Alfaro, Manuel Almeida, Angel Alonso, Agustin Alonso, Javier Alseda i Soler, Lluis Alsina, Claudi Alvarez, Manuel Amengual, Miguel Amoros, Jaume Ancochea, German Ansemil, Jose M Anton, Jose M Antonino, Jose A Ara, Pere Archilla, Bosco G Arias, Arturo F Arias-de-Reyna, Juan Arino, Miguel A Artal-Bartolo, Enrique Asensio Mayor, Jose Ayala-Gomez, Rafael Baeyens, Enrique

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (26 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Bagaria, Joan Balbuena, Luis Balibrea, Francisco Ballesta, Pascual N Banderas, Antonio Barcelo, Bartolome Barja, Jose M Barlabe, Antoni Barquin, Julian Barrera-Rosillo, Domingo Barrio, Roberto G Bartoll, Salud Bastero, Jesus Bayer, Pilar Bayod, Jose M Bellot, Francisco Belzunce Torregrosa, Felix L Benitez, Carlos Benito Munoz, Manuel Bermudez, Alfredo Bernis, Francisco Blasco, Fernando Bonet, Jose Bonilla, Luis L Branquinho, Amilcar Bravo De La Parra, Rafael Bru, Rafael Brunat, Josep M Bujalance, Emilio Burgos Gil, Jose I Busque, Claudi Cabrera, Antonio Cabrera-Garcia, Miguel Caceres, Jose Cachafeira, Alicia Calvino, Santiago Calvo, Manuel Camarena Badia, Vicente Campos, Javier Canadas-Pinedo, Maria A Candela, Vicente F Canto, Rafael Carro, Maria J Casacuberta, Carles Casas, Eduardo Castella Arteta, Ignacio Castellet, Manuel Castillo, Jesus M Cerda, Joan Ceron, Gaston Cifuentes, Patricio Climent Coloma, Joan J Comellas, Francesc Companys, R Conde, Juan M Contreras, Lucia Cordero, Luis A Corona, Carlos M Costa-Castello, Ramon Crespo, Teresa Crossley, Martin D file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (27 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Curbera, Guillermo P De La Fuente, Jose L de las Obras, Carmen de Leon, Manuel Dehesa, Jesus S del Bano, Sebastian del Rio, Roberto R Del Val, Pablo Delgado, Felix DeMiguel, Luis M Diaz, J I Aboites, Vicente Aceff-Sanchea, Flor d Acosta-Abreu, Roberto Aguilar, Marcelo A Alarcon, Hugo R Alberto, Lorandi M Aldama, Alvaro A Alonso, Alberto Arizmendi, Hugo P Arroyo, Maria-Jose Avila-Murillo, Fernando Avila-Vales, Eric J Betancourt, Carlos M Bosch, Carlos Brady, Wray G Bromberg, Shirley Caballero, Maria-Emilia Cadenillas, Abel Cambray-Nunez, Rodrigo Campos, Rafael G Campuzano, Gabriel Canavati, Jose A Cano-Garc/'es, Agust/i n J Carrillo, Humberto Carrillo Maria Melina, Becerra Casillas Macias, Andrea Castano, Victor M Cavero, Veronica Cervantes, Salvador Clapp, Monica Comer-Barragan, Enrique Corona-Corona, Gulmaro Cruz-Sampedro, Jaime de los Cobos Silva, Sergio Del Riego de Del Castillo, L Del Rio, Rafael R Diaz Gomez, Jose Luis Eckertson, Fred W Elizondo, E J Escarela Cornejo, Samuel Esparza, Sergio O Esquela, Chico Esquivel-Avila, Jorge A Esteva, Lourdes Estrada, Luis Eudave-Munoz, Mario Falconi-Magana, Manuel Fetter, Hans L Fuente-Florencia, Maria C

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (28 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Galaz-Fontes, Fernando Galeana-Sanchez, Hortensia Garcia, Gerardo E Garcia, Jesus E Garcia Almeida, Gerardo E Garcia Corte, Julio C Garcia-Moreno, Enrique E Garciadiego, Alejandro R Garibay, Fernando Garza-Garcia, Jose L Glazman, Mary Gomez, Campos G Gomez, Guillermo L Gomez, Javier A Gomez, Luis E Gomez, Susana Gonzalez, Hernan Gonzalez, Jesus Gonzalez, Raul E Gonzalez-Barrios, Jose M Gordienko, Evgueni I Gorostiza, Luis G Guillermo, Fernandez A Hector, Lara-Chavez Hennart, Jean P Hernandez-Garcia, Constancio Hernandez-Lerma, Onesimo Herrera, Ismael Hidalgo de Ita, Pedro Hinojosa-Berrones, J A Huerta, Jose 4 Idalia, Flores d Imaz, Carlos Ize, Jorge A Jazcilevich, Aron D Juan-Pineda, Daniel Kalicanin, Ivica Kravchenko, Vladislav V Lacomba, Ernesto A Lara-Aparicio, Miguel Lascurain Orive, Antonio Lizarraga, Carlos C Lluis, Emilio R Lluis-Puebla, Emilio Lopez-Mimbela, Jose f Lopez-Yanez, Alejandro Lugo-Goytia, Jose i Luque, Fernando V Madrid, Humberto Madrid Nunez, Bertha A Malo, Salvador Martinez-Morales, Manuel Arias-de-Reyna, Juan Bravo De La Parra, Rafael De La Fuente, Jose L de las Obras, Carmen de Leon, Manuel Dehesa, Jesus S del Bano, Sebastian del Rio, Roberto R Del Val, Pablo file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (29 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Delgado, Felix DeMiguel, Luis M Diaz, J I Diaz, Josep Diaz-Miranda, Antonio Diaz-Quinones, Miguel A Dominguez, Jesus M Donat, Rosa Gonzalez-Diez, Gabino Martin de Diego, David Palencia de Lara, Cesar Perez de Vargas, Alberto Quintanilla de Latorre, Ramon Sanchez Dehesa, Jesus Echeverria, Javier Elduque, Alberto Elias, Juan Escauriaza, Luis Escoriza-Lopez, Jose Escriba, Lourdes B Escudero, Laureano F Esteban, Juan R Rodriguez-Exposito, Jose Facenda Aguirre, Jose A Falco, Antonio Farre, Rafel Fernandez, Jose L Fernandez Moral, Emilio Fernandez Perez, Carlos Fernandez-Rodriguez, Marisa Ferrer, Jesus Floria, Luis Font, Josep M Freire, Emilio Martinez Finkelshtein, Andrei Ayala-Gomez, Rafael Burgos Gil, Jose I Cabrera-Garcia, Miguel Gadea, Pedro M Gadella, Manuel Gaeta, Federico Gago, Felipe Galacho, Cristobal R Gale, Jose E Galindo, Alberto T Gallego, Eduardo Galvez, Javier F Gandarias, Maria Garay, Oscar J Garcia, Domingo Garcia, Jose L Garcia-Alvarez, Victor O Garcia-Cuerva, Jose Garrido, Isabel Garriga, Ana J Gasca, Mariano Gil-Medrano, Olga Gimenez, Domingo Giraldo, Antonio Girela, Daniel

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (30 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Gomez, Joan V Gomez Pardo, Jose L Gomez-Ruiz, Francisco Gomis, Salvador S Gonzalez, Benito J Gonzalez, Manuel Gonzalez-Diez, Gabino Gonzalez-Lopez, Artemio Gonzalez-Vega, Laureano Gonzalo, Jesus Hernandez Gomez, Joaquin Martinez-Garcia, Vincente Pascual-Gainza, Pere Perez Gonzalez, Fernando Sanchez-Giralda, Tomas Santos-Garcia, Gustavo Haefner, Jeremy A Hayek, Nacere Herbera, Dolors Hermida-Alonso, Jose A Hernandez, Eugenio Hernandez, Fernandez T Hernandez, Jesus Hernandez, Rafael Hernandez, Salvador Hernandez, Vicente Hernandez Gomez, Joaquin Hernandez Ruiperez, Daniel Hernandez-Guerra, Juan M Hernandez-Penalver, Gregorio Herrero, Miguel A Herrero Ruiz, Francisco Hervas, Antonio Martinez Hernandez, Juan Alseda i Soler, Lluis Ibort, Luis A Ibort-Latre, Luis A Inforzato, Nelson F Rios-Insua, David Jara Martinez, Pascual Jimenez, Javier Jimenez, Jose M Jorba, Angel Kazarian, Kazaros S Bravo De La Parra, Rafael De La Fuente, Jose L de las Obras, Carmen de Leon, Manuel Ibort-Latre, Luis A Larrazabal, J M Larriba-Pey, Josep L Munoz-Lecanda, Miguel-C Palencia de Lara, Cesar Quintanilla de Latorre, Ramon Asensio Mayor, Jose Jara Martinez, Pascual Lopez-Marcos, Juan C Macias-Virgos, Enrique Maestre, Manuel Majadas, Javier Mancebo, Francisco J file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (31 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Manlleu, Josep G Marcellan, Francisco Marcos, J E Marhuenda, Francisco Marin Malave, Santiago Marin-Munoz, Leandro Marin-Solano, Jesus Marques, Francisco Marquez, Juan-Bosco R Marquez-Perez, Alberto Marquina, Antonio Martel, Antonio F Martin de Diego, David Martin-Guzman, Pilar M Martin-Reyes, Francisco J Martinez, Jose J Martinez Finkelshtein, Andrei Martinez Hernandez, Juan Martinez-Amores, Pedro Martinez-Garcia, Vincente Martinon, Antonio Mas Mari, Jose Massaguer, Josep M Mazon, Jose M Munoz Masque, Jaime Oriol Marti, Josep M Carrillo Maria Melina, Becerra Casillas Macias, Andrea Falconi-Magana, Manuel Madrid, Humberto Madrid Nunez, Bertha A Malo, Salvador Martinez-Morales, Manuel Martinez-Sanchez, Jorge Martinez-Villa, Roberto Renteria Marquez, Carlos Navarro, Gabriel Navarro, Joaquin P Navarro Sandalinas, Joaquim Naveira, Antonio M Nicolas, Jose A Nieto, Juan J Nitchiporenko, Alexandre I Noy, Marc Nualart, David Nunez, Marina Nunez, Pascual B Cambray-Nunez, Rodrigo Madrid Nunez, Bertha A Nunez-Doval, Ramon M de las Obras, Carmen Oleaga, Gerardo E Oriol Marti, Josep M Ortega-Cerda, Joaquim Otal, Javier Otero, Jesus Otero, Maria C Oubina, Jose A Lascurain Orive, Antonio Ojeda-Pena, Eduardo M

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (32 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Olive, Victor M Bravo De La Parra, Rafael Canadas-Pinedo, Maria A Fernandez Perez, Carlos Gomez Pardo, Jose L Hernandez-Penalver, Gregorio Hueso-Pagoaga, Jose L Larriba-Pey, Josep L Marquez-Perez, Alberto Mendez-Perez, Jose M Pacheco-Castelao, Jose M Palanques-Mestre, August Palencia de Lara, Cesar Pallares, Antonio J Pardo, Enrique Parra, Ignacio E Pascual, Griselda Pascual-Gainza, Pere Pascuas, Daniel Paul, Pedro J Paya, Rafael Penades, Jose Peral, Ireneo A Perez Carreras, Pedro Perez de Vargas, Alberto Perez Gonzalez, Fernando Pericas, Jordi Sanchez-Pedreno, Salvador Torres-Peral, Francisco J Trias Pairo, Joan Juan-Pineda, Daniel Morales-Perez, Jose L Ojeda-Pena, Eduardo M Pastor, Guilleimo Perez Romero, Jose J Perez-Chavela, Ernesto Perez-Rojas, Carlos Diaz-Quinones, Miguel A Quijada, Antonio Quintanilla de Latorre, Ramon Quintero, Antonio Quiros, Adolfo Quiroga, Raul Arias-de-Reyna, Juan del Rio, Roberto R Fernandez-Rodriguez, Marisa Martin-Reyes, Francisco J Mumbru Rodriguez, Pere Rafael, Lopez C Ramos, Juan I Ras, Antoni Reguera, Ana J Reyes, Miguel Riaza, Roman Rios-Insua, David Rodriguez, Fernando F Rodriguez, Gerardo Rodriguez, Sanjurjo J Rodriguez-Exposito, Jose Roig, Agusti Rojo, Jesus file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (33 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Sanchez-Reyes, Javier Viano Rey, Juan M Del Riego de Del Castillo, L Del Rio, Rafael R Mendoza Reyes, Miguel A Mendoza Rivera, Jose C Perez-Rojas, Carlos Raggi, Francisco F Ramos, Luis Recillas-Pishmish, Sevin Renteria Marquez, Carlos Rivera, Antonio F Rivera, Jose A Tapia-Recillas, Horacio Lopez-Sanchez, Jose M Montesinos Santalucia, Vicente Navarro Sandalinas, Joaquim Sabadell, Javier Saludes, Jordi San Miguel, Angel Sanchez, Juan M Sanchez Dehesa, Jesus Sanchez-Calle, Antonio Sanchez-Giralda, Tomas Sanchez-Pedreno, Salvador Sanchez-Reyes, Javier Sanchez-Umbria, Juan J Sansigre, Gabriela Santos, Francisco Santos-Garcia, Gustavo Sanz, Miguel A Sanz-Serna, J M Sanz-Sole, Marta Urbano-Salvador, Ana M Aceff-Sanchea, Flor d Cruz-Sampedro, Jaime Galeana-Sanchez, Hortensia Martinez-Sanchez, Jorge Mercado Sanchez, Gema A Saavedra, Patricia B Sabina, Federico J Salazar-Leyva, Enrique Sanchez, Hector F Sanchez-Garduno, Faustino Sanchez-Valenzuela, Oscar A Sautto Vallejo, Jose M Tovar S, Luis M Belzunce Torregrosa, Felix L Tarres, Joan Tena, Juan Torrecillas, Blas Torregrosa, Joan Torrens, Antoni Torres, M Torres-Peral, Francisco J Trias Capella, Rosa M Trias Pairo, Joan Trillas, Enric Tapia-Recillas, Horacio Tkachuk, Vladimir V

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (34 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Tornell, Jessica F Torres-Falcon, Yolanda Tovar S, Luis M Ulloa-Azpeitia, Ricardo Urias, Jesus Sanchez-Umbria, Juan J Udina, Frederic Urbano-Salvador, Ana M Avila-Vales, Eric J Martinez-Villa, Roberto Moreno-Valle, Javier F Sanchez-Valenzuela, Oscar A Sautto Vallejo, Jose M Valencia, Marco A Vallejo, Ernesto Vargas, Cristobal Vasilevski, N L Velasco-Hernandez, Jorge X Verde Star, Luis Vila-Freyer, Ricardo F Villarreal, Rafael H Villasana, Jose A Villegas-Silva, Luis M Walias, Magdalena Wecker, Segarra Welters, Gerald E Lopez-Yanez, Alejandro Yebra, Jose L Yguacel, Eugenio J Zaldivar, Felipe Zavadskij, Alexander G Zhevandrov, Peter N Zoreda-Lozano, Juan J ---------FRENCH Abadie, Jean M Airault, Helene Akbar-Zadeh, Hassan Akian, Jean-Luc Akian, Marianne Al Assaad, Al Ayham A Alain, Guimier Alarcon, Guy Albugues, Alain V Alinhac, Serge J Almeida, Luis N Amrouche, Denise Anantharaman-Delaroche, Claire Andler, Martin J Angeniol, Bernard Anker, Jean-Philippe Apkarian, Pierre Aribaud, F Arino, Ovide H Arnold, V I Arnoux, Pierre Artola, Michel Artzrouni, Marc Ashwin, Peter B Atrops, Sylvain Aubin, Thierry E file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (35 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Audin, Michele M Avez, Andrew Azzedine, Mohamed M Bacher, Roland Baillon, Jean-Bernard Baker, Mark D Balinski, Michel L Ball, Derek Balsalobre, Francois Bamberger, Alain Bardos, Claude W Barre, Raymond C Barriot, Jean Pierre Barsky, Daniel Bataille, Michel Batard, Eric Bauer, Max Bayer-Fluckiger, Eva Bayle, Lionel Beauville, Arnaud Becker, Jean-Marie Becker, Richard E Beedassy, Lekraj Bekka, Karim Bekka, Mohammed E Beliveau, Jean Bellissard, Jean V Arous, Gerard n Benoist, Yves Bensoussan, Alain Benzi, Michele Berger, Marcel Bermond, Jean C Bernadou, Michel Bernard, Alain Bernardi, Christine Berry, Gerard Bertrand, Daniel A Bertrand, Francois Bertrandias, Jean-Paul Bessis, Daniel J Besson, Gerard Bethery, Jean Bineau, Michel J Boeckle, Gebhard Bohnke, G C Bonami, Aline G Bonavero, Laurent Borgnet, Eric N Bossavit, Alain Bost, Jean-Benoit Bouleau, Nicolas Bourgeat, Alain P Bourguignon, Jean-Pierre Bourion, Georges Bourquin, Frederic L Boutillon, Jacques Boutot, Jean F Brasselet, Jean-Paul Breen, Lawrence

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (36 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Brenier, Yann Bretto, Alain Breuil, Christophe O Brezinski, Claude Brezis, Haim Briend, Jean-Yves Brisebois, Patrick Brodeur, Martin Broise, Michel J Broue, Michel Brown, Richard H Bruter, Claude P Bryant, John G Burnol, Jean-Francois H Cabane, Robert Caboz, Regis Cabre, Xavier Cabane, Robert Caboz, Regis Cabre, Xavier Cahen, Paul-Jean Caloz, Gabriel Camion, Paul F Carraro, Laurent Carrieu, Herve Cartan, Henri Cartier, Pierre Cathelineau, Jean-Louis Cauty, Robert Cegielski, Patrick Chabert, J L Chaperon, Marc Chapuis, O Charpentier, Marie R Charpin, Pascale Chasse, Guy Chatelin-Chaitin, Francoise H Chatelon, Jacques A Chauvin, Andre Chavent, Guy Chenciner, Alain Chipot, Michel Chone, Jacques Choquet, Gustave Choulli, Mourad Ciarlet, Philippe G Cibils, Claude Le Calvez, Patrice Anantharaman-Delaroche, Claire Dabaghi, Zakia B Damlamian, Alain Damphousse, Vincent Daniel, Marc Danset, Renaud Daouda, Sangare de Branges, Louis de Mesnard, Louis De Roeck, Yann H Deguenon, Charles Deheuvels, Paul Deheuvels, Rene file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (37 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

del Solar-Petit, Emilio Delange, Hubert Dellacherie, Claude Delporte, Jean Demers, Jacques Derriennic, Yves Desarmenien, Jacques Deshouillers, Jean-Marc Desolneux-Moulis, Nicole Detraz, Jacqueline Detrez, Eric L Dia Boun, Oumar Diarra, Bertin Dias, Frederic Dichi, Henri Diener, Marc Dierieck, Claude L Digne, Francois G Dincbas, Mehmet Dolbeault, Jean Dolbeault, Pierre E Gasqui De St-Joachim, Jacques Pham Dinh, Tao Van den Reysen, Joseph M Edixhoven, Sebastian J Ekeland, Ivar Ekong, Samuel El Jai, Abdelhaq El Kacimi Alaoui, Aziz Elbaz-Vincent, Philippe Elhadad, Jimmy Emamirad, Hassan Emery, Michel Epstein, Henri Escassut, Alain Esteban, Maria J Estivalezes, Jeanluc Euler, Reinhardt Eytan, Michael Bayer-Fluckiger, Eva Faille, Isabelle Faraut, Jacques Farina, Bruno Fathi, Albert Faure, Robert Faurre, Pierre L Fellah, Dominique Feray Beaumont, Margarida P Ferrif, Bernard Flajolet, Philippe Flato, Moshe Fliess, Michel Foata, Dominique Foulon, Patrick Fouque, Jean-Pierre Fraysse, Valerie Fresneau, Alain R Gabay, Jacques Gallardo, Luis Garcin, Jean-Louis

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (38 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Gasqui De St-Joachim, Jacques Gaudier, Henri Gauduchon, Paul Gaudy, Bruno Geoffrion, Bernard Genet, Jean R Gerard, Patrick Gerardin, Paul Gergondey, Robert A Germain, Emmanuel C Germain, Paul Gilbert, Jean-Charles Gilewicz, Jacek Gillard, Roland D Gilquin, Herve Ginestet, Jean-Paul Giraud, Luc Giraudet, Michele Godefroy, Gilles B Goffinet, Daniel Goichot, Francois Golse, Francois J Gondard-Cozette, Danielle J Gonnord, Michel Verger-Gaugry, Jean-Louis Habegger, Nathan Haddad, Labib S Hebey, Emmanuel Hedou, Veronique Heintz, Bruno Helein, Frederic Hellegouarch, Yves Hemard, Denis Henrot, Antoine Herman, Michael Herr, Laurent Indjoudjian, M D Iochum, Bruno Ivol, Bernard El Jai, Abdelhaq Gasqui De St-Joachim, Jacques Jaffard, Stephane P Jakubowicz, Daniel Jami, Adrien C Jouanolou, Jean-Pierre Juditsky, Anatolii B Lejeune-Jalabert, Monique Moser-Jauslin, Lucy El Kacimi Alaoui, Aziz Kahane, Jean-Pierre Kahn, Bruno Kaplan, Pierre Karoubi, Max Kassel, Christian Keller, Bernhard M Kern, Michel E Mokhtar Kharroubi, Mustapha La Shelle, Arthur G Labesse, J P Labrousse, Jean-Philippe Lacaze, B file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (39 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Lachaud, Gilles Lafitte, Olivier D Lafleur, Guy Lafon, Jean P Lafontaine, Jacques Lapeyre, Bernard Lascaux, Patrick M Lasry, J M Lassere, Patrice Laudenbach, Francois Laugier, Jean-Christophe Laumon, Gerard Laurencot, Philippe Laurent, Trilling Lazarus, Michel Le Calvez, Patrice Le Vey, Georges Leborgne, Daniel Lebourg, Gerard Lebret, Herve Leca, Pierre Leclerc, Bernard Lecomte, Pierre C Lecot, Christian Lefebvre, Jean E LeFloch, Philippe G LeGland, Francois Lehning, Herve Lehobey, Frederic Lejeune-Jalabert, Monique Lelievre, Gilbert Lelong, P J Lemaire, Jean-Michel Lemarechal, Claude Lemarie-Rieusset, Pierre-Gilles LeMehaute, Alain Lemieux, Claude Lentin, Andre Leray, Jean Lerner, Nicolas Leroide, Raymond Leroy, Andre G Lescot, Jack Lescot, Paul Lesieur, Leonce Letac, Gerard LeTallec, Patrick Levine, Harold I Levitt, Gilbert Libermann, Paulette L Lichnerowicz, Andre Ligozat, Gerard Y Magnin, Louis Mahey, Philippe Malliavin, Marie-Paule Marchand, Bernard Mardin, Arif Marichal, Gilles Marion, Jean A Marle, Charles M

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (40 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Martel, Philippe C Matet, Pierre Mattei, Jean-Francois Maugin, Gerard A Maynadier-Averous, Genevieve Mitterand, Francois Hogbe-Nlend, Henri Nadiras, Francois Nataf, Frederic Neveu, Jacques Nguyen, Thanh V Nicolas, Jean-Louis Nikolskii, N K Nitrosso, Bruno A Noot, Rutger Nuss, Philippe Ofman, Salomon Oliver, Robert A Olivier, Grange Olivier, Jean-Pierre Ollivry, Jean Pascal Orieux, Karine del Solar-Petit, Emilio Pajitnov, Andrei V Papasoglu, Panagiotis T Pardoux, Etienne C Paris, Luis Paszkiewicz, Pascault Patin, Jean-Marc Patra, Francois Paulin, Frederic Pavel, Monique L Pelletier, Jean-Jacques Penot, J P Periaux, Jacques Perrin, Dominique Perrine, Serge Perthame, Benoit Peters, Chris A Peyre, Emmanuel R Peyriere, Jacques Pham, Tuan D Pham Dinh, Tao Philippe, Bernard Picavet, Gabriel Piger, Camille Piger, Jean De Roeck, Yann H Lemarie-Rieusset, Pierre-Gilles Rabah, Rabah Rahbar-Rochandel, Hamid Rambavilas, C D Raoult, Anne Rav, Yehuda Reese, Roy C Regalia, Phillip A Reisz, Daniel Renault, Jean N Reversat, Marc Revuz, Daniel R Reznikoff, Iegor file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (41 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Richard, Maurice Richer, Stephane Riedi, Rudolf H Ritter, James Roberts, Jean E Roche, Claude Rofman, Edmundo Roger, Claude Roland-Prillecart, Marianne Roy, Patrick Kosmann-Schwarzbach, Yvette Sabatier, Pierre C Sabbagh, Gabriel Sabbah, Claude Sablonniere, Paul Saint-Loup, Bernard Saladin, Yves Salinier, Alain Salles, Maurice Saloff-Coste, Laurent Samuel, Pierre Sanderson, Yasmine B Saralegi-Aranguren, Martin E Saumont, Remi L Saut, Jean C Schafke, R Schapira, Pierre B Schatzman, Michelle Schiffmann, Gerard M Schloss, Jean-Baptiste H Schmitt, Jacques Schneiders, Jean-Pierre Colliot-Thelene, J L Teissier, Bernard Teller, Patrick Temam, Roger Tenenbaum, Gerald Terpolilli, Peppino Thera, Michel A Theuveny, Bertrand C Thibault, Jocelyn Thibon, Jean-Yves Thien, Noel Thom, Rene Thomas, Jean-Claude Thomas, Jean-Marie Thomas, Pascal J Tilouine, Jacques Tits, Jacques L Tortelier, Patrick Tougne, Pierre Tran, Huy Q Treibich, Armando Tronel, Gerard Trotman, David J Trubuil, Alain Ugalde, Edgardo Ulmer, Felix A Dubois-Violette, Michel Elbaz-Vincent, Philippe

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (42 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Le Vey, Georges Van den Reysen, Joseph M Vandeur, Jean-Luc Vaquie, Michel Varouchas, Jean Vauthier, Jacques J Vercken, Christine Verger-Gaugry, Jean-Louis Verjovsky, Alberto S Veron, Laurent Villalongue, Pierre Vilotte, Jean-Pierre Vuillermot, Pierre A Vuillon, Laurent Wagneur, Edouard Waldschmidt, Michel Walker, Martin Wang, Dongming M Weissler, Frederic B Williamson, Francis U Willien, Francois Wolfmann, Jacques Wurzbacher, Tilmann Yebbou, Johan YueChiming, Roger Y Akbar-Zadeh, Hassan Zaleski, Stephane Zanotti, Jean-Pierre Zara, Francois Zizi, Khelifa Zocca, Valentino ---------GERMAN Abels, Herbert Adelman, Dave Albeverio, Sergio A Albrecht, Ernst Albrecht, Gerald W Albrecht, Raphael Albright, Louis Aldinger, Juergen P Alefeld, Gotz E Aleis, Hans P Aleman, Alexandru Altmann, Klaus Amberg, Bernhard Ambos-Spies, Klaus an der Heiden, Uwe Andras, Varga Anger, Gottfried Apel, Thomas Armbrust, Manfred K Arnold, Ludwig Ast, Markus Aumann, Gunter Bach, Gunter Bachem, Achim Bader, Ralf Baer, Christian Bandt, Christoph Barakat, Aliaa file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (43 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Barakat, Mohamed Barner, Klaus Barner, Martin Bartsch, Thomas J Bartz, Dirk Batt, Jurgen O Bauer, Heinz Bauer, Stefan A Baumeister, Werner F Baumgarten, Jennifer Baur, Walter Becker, Eberhard Becker, Thomas Beekmann, Wolfgang Behncke, Horst Behrendt, Frank Bell, Wolfgang Bemelmans, Josef Benner, Peter Benz, Walter A Berding, Christoph F Berens, Hubert Berger, Robert W Bergmann, Arthur Bergweiler, Walter Bernert, Jan Berntzen, Detlef Berntzen, Rainer Bessenrodt, Christine Beth, Thomas Betsch, Gerhard Beutelspacher, Albrecht F Bieri, Robert Bierstedt, Klaus D Bitter, Ingmar Blum, Heribert Bock, Hans G Boehm, Michael Boehmer, Klaus W Boettcher, Albrecht Bohner, Martin J Boltje, Robert Bonk, Mario Boos, Johann Borchers, Hans W Borchers, Hans-Jurgen Borchert, Bernd H Borges, Rudolf Born, Jens Bornemann, Folkmar A Brandenburg, Harald Braun, Manfred Braun, Rudiger W Brieskorn, Egbert Brinkmann, Hans-Berndt Brokate, Martin Brommundt, Eberhard W Brosowski, Bruno H Bruck, Rainer Brueggemann, Wolfgang

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (44 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Bruening, Jochen Bruns, Winfried Brussee, Rogier Buchanan, Thomas Buhl, Hans-Jurgen Buhring, Wolfgang Bulirsch, Roland Z Bunse-Gerstner, Angelika Byrne, Catriona M Cabos, Christian E Calmet, Jacques Carlsson, Renate Carstensen, Carsten Chokoe, K F Christopeit, Norbert Cabos, Christian E Calmet, Jacques Carlsson, Renate Carstensen, Carsten Chokoe, K F Christopeit, Norbert Chudej, Kurt H an der Heiden, Uwe de Vries, Sven Decker, Eva Deimling, K Dellnitz, Michael Delvos, Franz J Dempe, Stephan Deninger, C Denzler, Jochen Deuflhard, Peter J Dick, Rainer Diener, Karl-Heinz Dietmair, Thomas Dietsch, Karl Dipper, Richard T Doering, Boro Doktorskii, Ruwel Y Dold, Albrecht E Donnell, William A Eggers, Bleicke Ehrich, Sven Eichhorn, Jurgen Eiermann, Michael Elsner, Ludwig Elstrodt, J Emmrich, Claudio Emmrich, Etienne Engel, Arthur Enss, Volker Entenmann, Walter Epkenhans, Martin Eppel, Dieter P Erber, Friedrich M Erle, Dieter H Erne, Marcel Ernst, Oliver G Eschenbach, Dieter Grosse-Erdmann, Karl-Goswin Nour Eldin, Hassan file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (45 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Faltings, Kai Fassbender, Heike Fekete, Otto Fekete, Sandor P Feldvoss, Jorg Felscher, Walter Felshtyn, Alexandr L Fenske, Christian Ferber, Thomas Ferebee, Ann S Fichter, Michael Files, Steven T Fischer, Bernd Fischer, Gerd O Fischer, H Fleishman, Charles Fleishman, Helen Floret, Klaus Forster, Klaus-Jurgen Frank, Detlef Frank, Michael Franken, Uwe Frauendiener, Jorg T Frick, Hans Friedrichs, Bernd Fritsch, Rudolf Frommer, Andreas Fuchssteiner, Benno Jeltsch-Fricker, Rita Bunse-Gerstner, Angelika Gackstatter, Fritz Gaertner, Ralf Gaier, Dieter Gamst, Jens Gaschutz, Wolfgang Gasser, Ingenuin Gekeler, Ernst-Ulrich Gentzsch, Wolfgang G Gerhardt, C Gerisch, Wolfgang Getzler, Ezra Ghanaat, Patrick H Giebermann, Klaus Glaeske, Hans J Gneiting, Tilmann J Gobel, Rudiger G Goes, Gunther W Gonska, Heinz H Goodman, Oliver A an der Heiden, Uwe Habetha, Klaus Hachenberger, Dirk Haemmerlin, Guenther Haffner, Tobias Hagen, Hans Hagen, Thomas Hahn, Joerg M Hamenstadt, Ursula Hanke, Martin Happel, Dieter

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (46 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Happle, Wolfgang Harborth, Heiko Harmand, Peter Harms, Torsten Hartl, Johann Hartler, Gisela R Harzheim, Egbert Hasenjaeger, Gisbert F Haussmann, Werner Hayer, Mattias P Heck, Matthias K Heil, Erhard Heilmann, Margareta Heiming, Helmut Heineken, Hermann Heinlein, Gerd Heintze, Ernst Heinze, Joachim Helm, Martin Helmer, Dietrich Helmke, Uwe R Helwig, Karl-Heinz Henk, Martin Henn, Hans-Werner Henney, Dagmar R Hermann, Martin Herrmann, Carsten H Herrmann, Norbert Herwig, Bernhard Herzberger, Juergen P Herzog, Gerd Herzog, Ivo Hettich, R Heusener, Michael Heuser, Harro G Heyer, Herbert K Hoffman, Roald Ihringer, Thomas Jacob, Niels Jaeger, Arno Jahn, Johannes Jahnke, Thomas Jainta, Paul Jannsen, U Jarre, Florian Jeggle, Hansgeorg Jeltsch-Fricker, Rita Jongen, Hubertus T Jost, Juergen K Junek, Heinz J Jungnickel, Dieter Kadison, Lars Kairies, Hans H Kalhoff, Franz B Kalmbach, Gudrun Kaniuth, Eberhard Kanold, Hans-Joachim Kanzow, Christian Karbe, Manfred J Karcher, Hermann Karpinski, Marek file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (47 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Kaul, Helmut Kauftail, Larry Kawohl, Bernhard Kegel, Otto H Keil, Detlef Keil, Karl-Heinz Keller, Wilfrid Keller, Wolfgang Kellerer, H G Kellerhals, Ruth Kemnitz, Arnfried Kerber, Adalbert Kerby, William E Kerner, Otto Kerns, Richard B Ketterl, Rudolf Kidwaii, Hariss Kiechle, Hubert Kimmerle, Wolfgang O Kings, Guido Kirchgassner, Klaus W Kirsch, Andreas Kiyek, K Krieger, Heindrich Landes, Hermann Lang, Bruno Langbein, Frank C Lange, Horst R Lanzinger, Hartmut X Laska, Michael Lass, Bodo Lasser, Rupert Latting, Charles R Laufmann, Charles Laufmann, Claire Laugwitz, Detlef Lauterbach, Reiner Leeb, Klaus Lefmann, Hanno Leinen, Felix Lempken, Wolfgang Lenk, Ursula Leptin, Horst A Leugering, Gunter R Liebers, Annegret C Liebers, Thorsten Lindemann, Willi Maennel, Hartmut Malkowsky, Eberhard Maltese, George J Mann, Matthias Manolache, Nicolae Marquardt, Wolfgang Marten, Wolfgang Martensson, Bengt K Martin, Reiner Marz, Roswitha Mathieu, Martin Matthies, Hermann G Mattner, Lutz

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (48 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Matzat, B H Maurer, Helmut Mayer, Dieter H Mayer, Guenter A Mayer, Karl H Mayr, Ernst W Mbaeyi, Peter N Nabben, Reinhard Nagel, Rainer Nastold, Hans-Joachim Neeb, Karl-Hermann Nelius, Christian-Frieder Nemethi, Csaba Neubrand, Michael Neunzert, Helmut Niessen, Heinz D Niethammer, Wilhelm Niggemann, Michael Noelle, Sebastian W Nolte, Nikolaus G Nonnenmacher, Dirk J Nour Eldin, Hassan Novak, Erich Nuessler, Thomas K O'Brien, Eamonn A Oberschelp, Walter Oeljeklaus, E Opfer, Gerhard H Ottmann, Thomas A Pallaschke, Diethard Pareigis, Bodo Pavlik, Norbert R Pawelke, Siegfried H Peisch, Heinz Peitgen, Heinz O Penzel, Frank Perisic, Vesna Peters, Alexander Peters, Meinhard H Petersen, Johannes A Petersson, H P Peyerimhoff, Alexander Pfau, Matthias Pfeiffer, Helmut Pflaum, Christoph Pflaum, Markus J Pflug, Peter Philip, Davaraj Picard, Rainer H Piepmeyer, Lothar Pier-Ribbert, Erwin Schulze-Pillot, R Quapp, Wolfgang Pier-Ribbert, Erwin Rack, Heinz-Joachim Rademacher, Hans-Bert Rannacher, Rolf C Ratschek, Helmut Rehmann, Ulf Reinschke, K J Reischuk, Rudiger K file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (49 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Reissel, Martin Reiszig, Gunther Remus, Dieter Rendall, Alan Render, Hermann Renner, Gerhard Reuding, Thomas Richert, Walter R Richter, Frank Richter, Guenther E Richter, Michael M Riede, Adolf Rieder, Andreas Rieger, Georg J Riemenschneider, Oswald Rigger, Ralf O Ringel, Claus M Roch, Steffen Roggenkamp, Klaus W Rohlfs, Jurgen Rohmfeld, Rainer F Van den Reysen, Joseph M Vasanta Ram, Venkatesa I Sachs, Ekkehard W Salzmann, Helmut R Sampson, Scott M Sander, Jurgen W Sandor, Viktor Sarangarajan, Aravamuthan Sarnes, Klaus-Dieter Sauvigny, Friedrich Schaal, Werner G Scharffenberger, Ulrich Scharlau, Rudolf M Schauenburg, Peter C Scheffler, Hans-Peter Scheiderer, Claus Schempp, Walter Schendel, Udo Schenzel, Peter Scheurle, Jurgen K Schiegl, Magda Schilling, Rene L Schimming, Rainer Schleicher, Dierk Schlichenmaier, Martin Schlichting, Gunter Schliep, Alexander Schluechtermann, Georg D Schmale, Wiland Schmalz, Gerd Schmekel, Maik Schmid, Peter P Schmidt, Roland Schmitz-Tewes, Axel F Schmudgen, Konrad Schneider, Gerhard J Schneider, Gunter Schneider, Manfred F Schneider, Rolf

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (50 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Schneiders, Robert J Schoemig, Ewald Schoenwaelder, Ulrich F Scholz, E Schomburg, Bernd Schon, Rolf Schottenloher, R M Schraegle, Horst Schrage, Georg Schrauf, Geza H Schroder, Johann Schroder, Johann Schroeder, Manfred R Schuch, Rudolf Schueller, George W Schuett, Werner Schuetze, Torsten Schultz, Rudiger Schulz, Friedmar J Schulz, Volker H Schulze-Pillot, R Schumann, Rainer Schutt, Ingo Schwachhoefer, Lorenz J Schwaenzl, Roland Schwaller, Thomas E Schwartz, Niels C Schwarz, Fritz Schwarz, Gunter Schwarz, Willi Schwermer, Joachim Schwetlick, Hubert Schwichtenberg, Helmut Scriba, Carl-Heinz Scriba, Christoph J Skinner, Seymour Westphal-Schmidt, U Schmitz-Tewes, Axel F Tabachnikov, Serge Takac, Peter Targonski, Gyorgy I Tasche, Manfred Tegtmeyer, Thomas K Teschl, Gerald Thalmaier, Anton Thedy, Armin Thiele, Ernst-Jochen Thieler, Peter Thoma, Elmar H Thoma, Manfred H Thorbergsson, Gudlaugur Tichatschke, Rainer Tillmann, Heinz G Tjiok, Mouw-Ching Tobiska, Lutz Topper, Jurgen Trautmann, G Trebels, Walter Trenkler, Goetz Troll, Gunter Ullrich, Peter file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (51 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Ulmet, Dan E Ulrich, Klaus Unger, Luise Upmeier, Harald Urbas, John de Vries, Sven van Hemmen, J L Van Trung, Tran Varnhorn, Werner Varsamis, Apostolos Vasanta Ram, Venkatesa I Veith, Christian P Velte, Waldemar Vetter, Udo Vogt, Dietmar Vogt, Elmar Voigt, Bernd Voit, Michael P Volkmann, Bodo Vollath, Hermann W Vollmers, Carsten von Grudzinski, Olaf Voss, Heinrich Waack, Stephen C Wagner, Dorothea Walcher, Sebastian Walter, Rolf W Walter, Wolfgang L Walter, Wolfgang V Walther, Hans-Otto Warnecke, Gerald Weck, Norbert Wefelmeyer, Wolfgang Wefelscheid, Heinrich Wegert, Elias Wehrhahn, Rodolfo Weidmann, Joachim Weidner, Jens Weidner, Peter Weiher, Jochen Weimar, Joerg R Weinert, Hanns J Welker, Volkmar Wells, Raymond O Wendland, Wolfgang L Wermuth, Edgar M Werner, Dirk Werner, Peter Werner, Wend Westphal-Schmidt, U Wette, E W Wiegmann, Klaus W Wielandt, Helmut W Wiesend, Gotz E Wild, Marcel Wildeshaus, Joerg W Winckler, Michael J Winkelmann, Joerg H Winkler, Jorg H Winter, Hermann

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (52 of 64) [19/1/2003 1:40:37]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Wisbauer, Robert Wissmann, Johannes W Wissner, Heinz-Wolfgang Witsch, Kristian Wolf, Hans-Diether Wolff, Manfred P Wolff, Roland Wuebbeling, Frank Wulff, Gerhard Yserentant, Harry Zamfirescu, Tudor Zapf, Hermann Zeidler, Eberhard Zeller, Karl Zeuner, Hansmartin Ziegler, Guenter M Ziegler, Martin Ziehms, Harald G Zillober, Christian Zimmer, Horst G Zimmermann, Alexander Zimmermann, Bernd Zimmermann, Irene A Zimmermann, Uwe T Zink, Ernst-Wilhelm Zink, Thomas F Zoerner, Thorsten H Zwick, Daniel S ---------GREEK Achilleas, Sinefakopoulos Akrivis, Georgios Alexopoulos, Vaggelis I Anagnostopoulou, Niki M Andreadakis, Stylianos Antoniadis, Jannis A Artemiadis, Nicolas Arvanitopoulos, George Arvanitoyeorgos, Andreas T Athanassopoulos, Konstantin Athanassoulis, Gerassimos Atherton, Stephen W Avgerinos, Evgenios P Axtsalotidis, Christoforos Bolis, Theodore S Boudouvis, Andreas G Bozonis, Petros Bredimas, Antoine P Camouzis, Elias Charalambous, Michael G Charalampos, S T Chryssakis, Thanassis Coletsos, Ioannis T Cotsakis, Spiros Cotsiolis, Athanase A Daras, Nicholas J Dassios, George Delistathis, George Demetriou, Ioannes C Diamantakis, Michail T Dimitracopoulos, Constantinos file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (53 of 64) [19/1/2003 1:40:38]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Dionysiou, Demetrios Djoganopolous, Christatos Dougalis, Vassilios A Drossos, Constantinos A Economou, Antonis Evagelopoulos, George Evangelatos, Dionyssis S Famelis, Ioannis T Famelis, T I Farmaki, Vasiliki A Fellouris, Anargyros G Flytzanis, E Fragoulopoulou, Maria Fragulis, George F Gallopoulos, Efstratios Georghiou, Constantinos Georgiou, D A Georgiou, Dimitris Georgiou, John C Gerontidis, Ioannis I Giannaros, Sakis G Gidiotis, Grigorios Grapsa, Theodoula N Gryllakis, Constantinos Hadjinicolaou, Maria E Hadjisavvas, Nicolas Hadjiloucas, Vasilis Haralampidou, M M Hartonas, Chrysafis Hatzinikolakis, Emmanuil Johnson, K G Kalfa, Kornilia S Kalogerakis, Giannis Kalpazidou, Sofia Karampetaxis, Nicholas P Karanasios, Sotirios Karanikas, Constantinos Karydas, Nicholas G Katavolos, Aristides Katsaounis, Theodoros Katsoprinakis, Emmanuel Kechagias, Nondas E Kehayopulu, Niovi Kioustelidis, John Kiriakie, Kiriaki Kirousis, Lefteris M Klaudatos, Nicos S Kobotis, Apostolos Koumboulis, Fotis N Kourouniotis, Christos Koutsopoulos, Constantine J Kouvidakis, Alexandros E Kravvaritis, Dimitrios C Krikelis, Peter Kydoniefs, Anastasios Kyriazis, Athanasios Lagos, Gerasimos Lambrinos, Panos T Leoussis, Paul C Louiza, Paschon

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (54 of 64) [19/1/2003 1:40:38]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Loukakis, Emmanuel Louridas, Sotiris Lyberopoulos, Athanasios N Makrakis, George N Maragkakis, Manolis G Markopoulou, Maria Marmaridis, Nikolaos P Mattheos, Adamantios S Mavrommatis, Kostas Melas, Antonios D Meletiou, Gerasimos Makrakis, George N Maragkakis, Manolis G Markopoulou, Maria Marmaridis, Nikolaos P Mattheos, Adamantios S Mavrommatis, Kostas Notaris, Sotirios Olum, Paul Panagiotopoulos, Pan D Pantelelis, Nikos G Pantelias, Anastasia Pantelidis, Georgios Papadaki, Malvina M Papadakis, Manos Papadimitrakis, Michael Papadopoulos, Basil K Papadopoulos, Panagiotis Papageorgiou, Nikolaos S Papakostas, Sotirios N Papaschinopoulos, Garyfalos Papathanassiou, George Papistas, Athanassios Paradis, Nikolas Pavlopoulos, Harry Perros, George Petalas, Chrysostom Petridis, Stacey Pheidas, Thanases C Phili, Christine M Philos, Christos G Poulas, Maria Rassias, Themistocles M Sakkalis, Takis P Sarantopoulos, Ioannis C Sarantopoulos, Yannis C Schinas, John Tertikas, Achilles Terzakis, Dimitris Thanos, Demosthenis A Theodoridis, Marinos G Theohari-Apostolidi, Theodora Thoma, Apostolos Valougeorgis, Dimitris Vardulakis, Antonis Varverakis, Andreas E Vassiliou, Efstathios Vetsis, Costas G Vlachakis, Ioannis S Voreadou, Rodiani Vougiouklis, Thomas N file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (55 of 64) [19/1/2003 1:40:38]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Vrahatis, Michael N Xenos, Philippos J Yeyios, Apostolos K Zahariades, Michael ---------INDIAN Adhikari, Sukumar D Agarwal, Girdhar G Aggarwal, Ravinder K Aggarwal, Satish K Agnihotri, Shripad G Agrawal, Bhagwan D Ahuja, Gopi Ajmal, Naseem Akerkar, Rajendra A Alphonse, A M Anandani, P V Arora, S K Athavale, Ameer Bahuguna, Dhirendra Bala, Shashi Balachandaran, Vadekke K Balachandran, K Balakrishnan, Rangaswami Banerji, Pradeep K Barua, Rana Basu, Adhir K Bhanu Murthy, T S Bhargava, S Bhaskara Rao, K P S Bhatia, Rajendra Bhattacharjee, Guru P Bhattacharya, Tilak Bhattacharyya, Pulin K Biswas, Dipankar Borkar, Vivek S Bose, Arup Bose, Ramendra K Bose, Sujit K Chakravarti, Nilotpal Chakravarti, R S Chandra, Peeyush Charak, Kuldeep S Chaudhary, B D Chhabra, Harmeet K Choudum, Sheshayya A Dani, S G Das, Shawpawn K Datta, Sunil Deo, Sadashiv G Deo, Satya Deshpande, Vishwanath L Dhar, Joydip Dutta, Mahadev Dwivedi, Shankar H Fakhruddin, Syed M Fernandes, Olga J Gadam, Sudhasree Gautam, Vishvajit Vikram S Gebermedhen, Kifle

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (56 of 64) [19/1/2003 1:40:38]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Ghorpade, Sudhir R Ghosh, Parimal K Goyal, Krishan C Gupta, Dharma P Gupta, Kailash C Gupta, Manjul Gupta, Ram K Gupta, Ram N Gupta, Renu Gupta, S P Gupta, Satya N Gupta, Subhash C Gupta, Vijaya P Hans-Gill, Rajinder J Hanumanthachari, Juturu Harinath, K S Hegde, S M Hussain, T Jagadeeshan, Shashidhar Jain, Jainendra K Jaiswal, U C Janakiraman, S Jayanthan, A J Jayaram, S R Joshi, Chandra M Kadalbajoo, Mohan K Kalpakam, S Kandaswamy, P Kannan, V Kanta, Ravi Kapoor, Om P Karia, Dinesh J Kesavan, Srinivasan Keskar, Pradipkumar H Khanda, Mahati Konnully, Augustine J Krishna, J G Krishnamurthy, Visvanatha Kulkarni, Sudhir H Kumar, Ashish Kumar, Ashok Kumar, Naveen Kumar, Vinod Kumartyagi, Vinod Kumbhat, R K Kundu, Subiman Lahiri, B K Laipubam, Gopal S Lahiri, B K Laipubam, Gopal S Lakshmanan, Muthusamy Lal, Shiva N Limaye, Balmohan V Mahanti, Anirban Mukerjee, Indranil Nag, Debashis Nag, Subhashis Naidu, K B Nair, M T Nambooripad, K S Nanda, Vinod C file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (57 of 64) [19/1/2003 1:40:38]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Nandakumaran, A K Narendra, Nayak N Narendran, G M Natarajan, Ponnammal Nath, Vishwa Nirmaljeet, Kaur Pachpatte, Baburao G Pachpatte, Deepak B Pagey, Subhash S Pandeya, Bashishth M Pani, Amiya K Paramhans, S A Parhi, Narahari Parida, Jayaram Parimala, Raman Parthasarathy, Kalyanpuram R Parthasarathy, Krishnan Parthasarathy, Panamalai R Parvatham, Rajagopalan Passi, Inder B Pati, Vishwambhar Patidar, Prakash C Patnaik, Surendra-Nath Bhaskara Rao, K P S Raghavendra, V Rai, Kabindra N Rai, Lajpat Raina, Ravinder K Raj, C P Rajan, B S Rama-Murthy, Kavi Ramakrishna, H Ramakrishnan, B Ramasubramanian, Sundareswaran Rana, Inder K Ransal, Ashok Rao, B L Rao, Gandikota L Rao, K C Rao, Taduri S Rathie, A K Reddy, G L Reddy, N G Sahadevan, Ramajayam Saksenan, Krishna M Salaria, Devi Singh Sarkhel, D N Sastry, K R Saxena, Ram K Tamizhmani, K M Tewari, Udai B Thakare, Nimba K Thandapani, E Thangaraj, V Thangavelu, Sundaram Thothathri, Venugopal Thrivikraman, T Tiwari, Ram N Tripathi, Amitabha Tripathi, Mukut M

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (58 of 64) [19/1/2003 1:40:38]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Vanninathan, Muthusamy Vasavada, Hema Vasavada, Mahavirendra H Vasudevan, T C Venkataraman, S Venkatesh, T Venkateswaran, S Verma, J K Verma, Meenu Verma, Rakesh Vermani, Lekh R Vidyasaqar, Mathukumalli Vijayakumar, Ambat Yadav, Sarjoo P Zacharias, Cecily ---------RUSSIAN Abakumov, Alexander I Adrianova, L Y Adukov, Victor M Agrachev, Andrei A Agranovich, M S Airapetyan, Ruben G Akhiezer, D N Akhmetiev, Peter M Aksentev, L A Aleksandrov, Aleksandr G Aleksandrov, Alexei B Alenitsyn, Alexander Aleshkov, Yuri Z Alexeevski, Andrei Alferov, Gennady V Aminova, A V Anan'ev, Boris I Andronov, Ivan V Anikonov, Yuri Anisov, Sergey S Antonets, Mikhail A Antropov, Alexander A Arslanov, Marat M Artamonov, V A Asekritova, Irina U Askhabov, Sultan N Avdonin, S A Avkhadiev, Farit G Azizov, T Y Bacurov, Vadim G Bagaev, Gennadiy N Bahturin, Yuri A Bakushinskii, Anatolii B Baranov, A S Basin, Michael V Baskakov, A G Baskakov, V A Batukhtin, Valentin D Bauer, Svetlana M Bazhenov, Maxim V Bekker, Boris Belegradek, Oleg V Belonogov, V A Belov, Evgenii G file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (59 of 64) [19/1/2003 1:40:38]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Belyankov, A Y Berezhnoi, Eugeniy I Berezin, Sergey A Berezovskaya, Faina Berkovich, Lev M Bessonov, Vladimir A Bibikov, Yuri N Bilgildeyeva, Tatiana Y Bilopolska, Ya I Blizorukov, Michail G Bloshanskii, Igor L Blyumin, Sam L Boguslavskii, L B Boikov, Ilja V Bokut, Leonid A Bolibrukh, Andrey Borisov, Igor S Borisovich, Yu G Borodin, Oleg V Botchev, Mikhail Bratus, Alexander S Bruno, Alexander D Brutyan, Murad A Brykalov, Sergei A Buchbinder, Ioseph L Buchstaber, Victor M Budkin, Alexander I Bugaenko, Vadim O Bukharaev, Rais G Bukhvalov, A V Bulgakov, Alexsander I Burd, Vladimir S Bure, Valeri Burnashev, Marat V Burov, Alexander Chebotarev, Alexander Chebotarev, Vladimir I Chentsov, Alexander Cherednichenko, Victor G Chernousko, Felix L Chistjakov, Sergei V Chistyakov, Vyacheslav V Danilaev, Peter G Danilov, Vladimir G Dekhtyar, Michael I Demidenko, Gennadii V Demidov, Sergei S Demidovitch, Vassili B Demyanov, Vladimir F Dmitrieva, L A Dmitruk, Andrei V Dobrokhotov, Serguei Y Dol'nikov, Vladimir L Dubinin, Vladimir N Dubovskii, Pavel Dubrovskii, V V Duzhin, Sergei V Edelstein, Sergej L Egorov, Alexandr A Egorychev, Georgy P

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (60 of 64) [19/1/2003 1:40:38]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Eiderman, Vladimir Y Elizarov, Alexander M Elizarov, Paul M Elkin, V I Eremin, Yu A Ermakov, Sergey M Ermolaev, Juri B Ershov, Eugene K Faminskii, A V Fedoryaev, Sergey T Federov, Sergei Fetisov, Slavi Filimonov, Andrey M Filippova, T F Fomin, V N Fonarev, Anatolii A Galiullin, Iljas A Galkin, Valery A Galkin, Vladimir M Gaposhkin, Vladimir F Gelfand, Mikhail S Gizatullin, Marat K Glavatsky, Sergei T Glazyzin, Vasiliy V Gliklikh, Yu E Glushenkov, V D Gokhman, Alexei O Goltjapin, Victor V Golubeva, Valentina A Golubov, Boris I Golubyatnikov, Vladimir Gomonov, Sergej A Helemskii, Alexander Y Igoshin, Vladimir A Ilyashenko, Yulij S Islamov, Galimzian G Ivanov, Evgeny A Ivanov, Sergei A Ivochkina, N M Izrailevich, Yakov A Juriev, Denis V Kabanov, Sergey N Kabanov, V V Kachurovskii, Alexander G Kalaidzidis, Yannie L Kalashnikov, Vladimir Y Kalyabin, Gennadiy A Kamenskii, Georgii A Karapetyants, Nicolai K Karasev, Mikhail Karol, Andrew Karpov, Yu G Karzanov, Alexander V Kasparitis, Darius Kasyanov, Victor N Katanova, Anna Kazarin, Lew S Kelzon, Andrei A Kerimov, Alexander K Kharshiladze, Alexander F Khavinson, Semen Y file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (61 of 64) [19/1/2003 1:40:38]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Khonkin, Alexander D Kireeva, Alexandra V Kirjanen, Alexander Koslov, A Kovalev, Alexei Labsker, L G Ladyzhenskaya, O A Landis, Eugenii M Larin, Alexander A Latkin, Eugene I Lavrentiev, Mikhail M Lebedev, Leonid P Leontovitch, Andrey M Lesokhin, M M Levchuk, Vladimir M Levendorskii, Sergei Z Levenshtam, Veleriy Lifanov, Ivan K Lifshits, Mikhail A Maergoiz, Lev S Magnitskii, Nicolay A Makhnev, Alexander A Makienko, Peter M Maksimov, Vladimir P Malishevski, Andrey V Malakhov, Vladimir Maloshevskii, S G Maryukov, Michael N Maslov, Leo A Matasov, Alexander I Matiyasevich, Yuri Matveev, Sergej V Matvejchuk, Marjan S Mazurov, V D Nagaev, S V Nakhman, Alexander D Napalkov, Valentin Nasyrov, Samyon R Nazaikinskii, Vladimir E Nepomnyashchikh, Yuri V Nesterov, A V Netsvetaev, Nikita Y Nezhmetdinov, I R Nikitina, Tatyana Nikolaevich, Kasyanov V Nikulin, Viacheslav V Noskov, Gennady Novikov, Igor Y Novikov, Sergei Y Novitskii, Igor M Nuzhin, Ja N Obrezanova, Olga A Obukhov, Valery V Obukhovskii, Valerii V Odinokov, O V Ohezin, Sergei P Oleinik, Olga A Onishchik, A L Osilenker, Boris P Osipenko, K Y

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (62 of 64) [19/1/2003 1:40:38]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Osipov, Edward P Ostapenko, Vladimir V Ovchinnikov, Peter G Ovseevich, Alexandre J Pakshin, Pavel V Pakuliak, Stanislav Z Palamodov, Victor Panov, Aleksandr N Panteleev, Andrei V Pantyukhina, Marina N Parinov, Ivan A Pasashkov, Sergei A Pavlenko, Vyacheslav N Persits, David B Pestov, G G Petrogradsky, Victor M Petrosjan, Leon A Petrov, Nikolai N Petukhov, Alexander Pevnyi, A B Pilidi, Vladimir S Pilyugin, Sergei Y Popovic, Peter Rabinovich, V S Raikher, Yurii Rapoport, Lev B Rappoport, Juri M Razzhevaikin, Valery N Remeslennikov, Vladimir N Rucinsky, Martin Sabinin, Lev V Salii, Viacheslav N Samokhin, Alexey V Samokhin, Mikhail V Sapozhenko, A A Sbitneva, Larissa Krasichkov-Ternovskii, Igor F Taimanov, Iskander A Taitslin, Mikhail A Tarasenko, Felix P Tarassenko, Felix P Terentiev, Alexej G Tersenov, S A Timofeev, Alexei Y Timofeev, Nikolai M Timofeev, Vladimir G Tkhai, Valentin N Tolstykh, Andrei I Tolstykh, Vladimir A Tovstik, Petr E Trofimov, Valerii V Trofimov, Vladimir I Uglanov, Alexei V Uspensky, Vladimir A Vakulenko, Alexander V Vakulenko, Sergei A Vasilchik, M Y Vasilev, Alexander Vasilev, Valery A Vassilevich, Dmitri V Vassiliev, Victor A file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (63 of 64) [19/1/2003 1:40:38]

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N

Vdovin, Viktor V Vedernikov, Valery K Venets, Ekaterina V Venets, Vladimir I Veretennikov, Alexander Y Verevkin, Andrew B Vershik, A M Vinberg, Ernest B Vishik, Mark Vishnevskii, Mikhail P Vladimirov, Andrey V Vodopyanov, S K Volodarskii, Alexander I Volodin, I N Vorobev, Oleg Y Voronin, Serguei M Voronov, Theodore T Voskresenskii, Valentin E Yagola, Anatoly Yakubov, Askhab Y Yaroshchuk, Igor O Yasnitsky, Leonid Yudin, Alexsandr A Yukalov, Vyacheslav I Yurinsky, Vadim Yurko, Vjacheslav A Zadorozhnii, Vladimir G Zak, Fyodor L Zakalyukin, Vladimir M Zaslavsky, Boris G Zelikin, M I Zharinov, Victor V Zhidkov, Peter E Zhislin, Grigorii M Zhitomirskiy, Grigoriy I Zhubr, Alexei V Zhukov, Igor B Zhuravlev, Vladimir G Zilber, Boris I Ziza, O A Zolotykh, Andrej A Zubov, Sergey V Zudin, Andrey N Zuev, Yuri A Zupnik, Boris M Zvonilov, Victor I ##########

file:///E|/WEBSITE/BIBLIO/TESTDATA/PEOPLE_N (64 of 64) [19/1/2003 1:40:38]

SimPack/Sim++ Simulation Toolkit

SimPack/Sim++ Simulation Toolkit SimPack is a collection of C++ tools (routines and programs) for computer simulation. The purpose of SimPack is to provide people with a "starting point" for simulating a system. The intention is that people will view what exists as a template (or "seed") and then "grow a simulation program." SimPack tools have been used for six years of teaching computer simulation at the under- graduate (senior) and graduate levels. Also, many people have taken the basic tools and created more complex research applications with lots of options specific to a particular area. Sim++ is a library of routines for implementing discrete event simulations, built by Robert Cubert and Paul Fishwick of the University of Florida. Priority queues are integral to such simulations, and Sim++ contains implementations of linked, binary, leftist, and calendar heaps. If you need a priority queue to control a simulation, check out ftp://bikini.cis.ufl.edu/oub/simdigest/tools . It also contains random number generators for a variety of different distributions, including uniform, exponential, and normal. Fishwick's book Simulation Model Design & Execution: Building Digital Worlds describes model design using SimPack.

● ●

Link to SimPack/Sim++ Download Files (local site)

Problem Links ● ●

Priority Queues (7) Random Number Generation (7)

About the Book Send us Mail Go to Main Page This page last modified on Apr 10, 1997.

file:///E|/WEBSITE/IMPLEMEN/SIMPACK/IMPLEMEN.HTM [19/1/2003 1:40:39]

Fire-Engine and Spare-Parts String and Language Algorithms

Fire-Engine and Spare-Parts String and Language Algorithms Fire Engine and Spare Parts -- Regular Language Algorithms FIRE Engine is a finite automata toolkit, written in C++ by Bruce Watson. It provides production quality implementations of finite automata and regular expression algorithms, specifically the construction of finite Several finite automata minimization algorithms have been implemented, including Hopcroft's O(n lg n) algorithm. Both deterministic and non-deterministic automata are supported, and it has been used for compiler construction, hardware modeling, and computational biology applications. It is strictly a computing engine, and does not provide a graphical user-interface. SPARE Parts is a string pattern recognition toolkit, written in C++ by Bruce Watson. It provides production quality implementations of all major variants of the classical string matching algorithms for single patterns (both Knuth-Morris-Pratt and Boyer-Moore) and multiple patterns (both Aho-Corasick and Commentz-Walter). Greatly improved commercial versions of both codes are available from www.RibbitSoft.com, with older versions available from www.RibbitSoft.com/research/watson/ and available by anonymous FTP from ftp.win.tue.nl in the directory /pub/techreports/pi/watson.phd/. Neither version in the public domain, and neither is copy-left. They are both freely available for noncommercial use (research, experimentation, etc., but not even shareware). They must not be redistributed, but people can simply tell their friends where to get their own copy.

● ● ● ● ● ●

Ribbit Software Systems (commercial versions) Bruce Watson's WWW Page" FTP to Original distribution HTML Distribution Email to Bruce Watson Download Files (local site)

Problem Links ● ●

Finite State Machine Minimization (8) String Matching (7)

file:///E|/WEBSITE/IMPLEMEN/WATSON/IMPLEMEN.HTM (1 of 2) [19/1/2003 1:40:40]

Fire-Engine and Spare-Parts String and Language Algorithms

About the Book Send us Mail Go to Main Page This page last modified on Apr 23, 1997.

file:///E|/WEBSITE/IMPLEMEN/WATSON/IMPLEMEN.HTM (2 of 2) [19/1/2003 1:40:40]

Algorithms in C++ -- Sedgewick

Algorithms in C++ -- Sedgewick

Sedgewick's popular algorithms text Algorithms in C++ comes in several different language editions, including C, C++, and Modula-3. It distinguishes itself through the use of algorithm animation, and in its broad topic coverage, including numerical, string, and geometric algorithms. The language specific parts of the text consist of many small code fragments, instead of full programs or subroutines. Thus they are best used as models rather than full implementations. The program fragments from the C++ edition have been made available from http://hegschool.aw.com/cseng/authors/sedgewick/algo-in-c++/algo-in-c++.html

● ●

Book homepage with algorithm fragments Sedgewick's own Algorithms and Data Structures Course

Problem Links ● ● ● ● ● ● ● ● ● ● ●

Sorting (5) Robust Geometric Primitives (4) Range Search (4) String Matching (4) Arbitrary Precision Arithmetic (3) Minimum Spanning Tree (3) Random Number Generation (3) Topological Sorting (3) Discrete Fourier Transform (2) Linear Programming (2) Matching (2)

About the Book Send us Mail Go to Main Page This page last modified on Apr 10, 1996. file:///E|/WEBSITE/IMPLEMEN/SEDGEWIC/IMPLEMEN.HTM [19/1/2003 1:40:41]

Geolab -- Computational Geometry System

Geolab -- Computational Geometry System

Geolab is a library of 2-D geometric algorithms and data structures, with a visualization environment, written by Pedro J. de Rezende, Welson R. Jacometti, Cesar N. Gon, and Laerte F. Morgado, Universidade Estadual de Campinas, Brazil. Geolab is written in C++ and requires the SUN C++ compiler, but a Sparc binary is included along with all source code. Geolab appears to be primarily for the brave, since it contains relatively little documentation, but it does contain 40 algorithms, including all the computational geometry classics plus some other interesting things like: ● ● ●

Farthest point Voronoi diagram Nearest neighbors Ray shooting

The Geolab distribution is available for ftp at the Univ. of Minnesota Geometry Center.

● ●

Geolab mirror distribution at the Geometry Center Download Files (local site)

Problem Links ●

Robust Geometric Primitives (5)

About the Book Send us Mail Go to Main Page This page last modified on Apr 23, 1996.

file:///E|/WEBSITE/IMPLEMEN/GEOLAB/IMPLEMEN.HTM [19/1/2003 1:40:41]

Grail: finite automata and regular expressions

Grail: finite automata and regular expressions Grail, by Darrell Raymond and Derick Wood, is a symbolic computation environment for finite-state machines, regular expressions, and other formal language theory objects. Using Grail, one can input machines or expressions, convert them from one form to the other, minimize, make deterministic, complement, and perform many other operations. Grail is intended for use in teaching, for research into the properties of machines, and for efficient computation with machines. Grail is written in C++. It can be accessed either through a process library or through a C++ class library. It can handle machines with 100,000 states and dictionaries of 20,000 words. All code and documentation is accessible from the WWW site http://www.csd.uwo.ca/research/grail. Version 2.5 of Grail enables you to manipulate parameterizable finite-state machines, parameterizable regular expressions, and parameterizable finite languages. By `parameterizable', they mean that the alphabet is not restricted to the usual twenty-six letters and ten digits. Instead, all algorithms are written in a type-independent manner, so that any valid C++ base type and any user-defined type or class can define the alphabet of a finite-state machine or regular expression.

● ●

GRAIL home page Download Files (local site)

Problem Links ●

Finite State Machine Minimization (9)

About the Book Send us Mail Go to Main Page This page last modified on Apr 29, 1996.

file:///E|/WEBSITE/IMPLEMEN/GRAIL/IMPLEMEN.HTM [19/1/2003 1:40:42]

Calendrical Calculations

Calendrical Calculations In two papers, N. Dershowitz and E. Reingold of the University of Illinois provide a uniform algorithmic presentation of a variety of different calendars systems, including the Gregorian, ISO, Julian, Islamic, and Hebrew calendars, as well as other calendars of historical interest. They provide Common Lisp and C++ routines to convert dates between calendars, day of the week computations, and the determination of secular and religious holidays. The implementations of these calendars and the papers describing them are available on our local distribution site.

● ●

Reingold's Calendar Page Download Files (local site)

Problem Links ●

Calendrical Calculations (10)

About the Book Send us Mail Go to Main Page This page last modified on May 8, 1996.

file:///E|/WEBSITE/IMPLEMEN/REINGOLD/IMPLEMEN.HTM [19/1/2003 1:40:43]

LINK -- Programming and Visualization Environment for Hypergraphs

LINK -- Programming and Visualization Environment for Hypergraphs LINK is a software system designed to be a general-purpose, extendible computing environment in which discrete mathematical objects representing real world problems can be easily manipulated and visualized. The system (which is still under development) features a full Scheme interpreter with access to the Tk graphics toolkit (STk), a flexible GUI, and a rich variety of important Collection and Graph objects which are grouped into C++ libraries. Link is designed to be useful as an educational tool, as a research tool, and as a prototyping tool for industry. The primary investigators of LINK are: Nathaniel Dean of Lucent / Bell Laboratories, Mark Goldberg of Rensselaer Polytechnic Institute, Greg Shannon of Spanning Tree Technologies, and Steven Skiena of SUNY Stony Brook. The project is currently coordinated by Jonathan Berry, a postdoctoral fellow at DIMACS at Rutgers University and one of LINK's primary designers. The LINK system is freely available for teaching and research. Also within the distribution directory is Jon Berry's implementations of several graph partitioning heuristics, including Kernighan-Lin, Simulated Annealing, and Path Optimization.

● ● ●

Download Files (local site) LINK distribution site at DIMACS Jon Berry's graph partitioning heuristics, original site

Problem Links ● ● ●

Graph Partition (8) Graph Data Structures (6) Set Data Structures (4)

About the Book Send us Mail Go to Main Page This page last modified on Dec 18, 1996.

file:///E|/WEBSITE/IMPLEMEN/LINK/IMPLEMEN.HTM [19/1/2003 1:40:44]

David Eppstein's Knuth-Morris-Pratt Algorithm and Minkowski sum code

David Eppstein's Knuth-Morris-Pratt Algorithm and Minkowski sum code David Eppstein's WWW page http://www.ics.uci.edu/~eppstein/161/kmp/ contains an implementation of the Knuth-Morris-Pratt string matching algorithm in C++. It exists as a teaching example from his algorithms course notes. He also provides a Mathematica code for the Minkowski sum problem in http://www.ics.uci.edu/~eppstein/junkyard/ukraine/ , which computes Minkowski sums of line segments (aka zonotopes). This problem is also closely related to maintaining line arrangements. The program works in any dimension, but doesn't do Minkowski sums of more complicated shapes.

● ● ●

Link to EPPSTEIN's Knuth-Morris-Pratt Algorithm Link to EPPSTEIN's Minkowski sums code in Mathematica Download Files (local site)

Problem Links ● ●

Minkowski Sum (4) String Matching (4)

About the Book Send us Mail Go to Main Page This page last modified on Feb 17, 1997.

file:///E|/WEBSITE/IMPLEMEN/EPPSTEIN/IMPLEMEN.HTM [19/1/2003 1:40:44]

GraphViz -- graph layout programs

GraphViz -- graph layout programs GraphViz is a program yielding polyline drawings of graphs. An email graph-drawing server is also available, if you would like to experiment with the program without downloading it. ● ● ●

Link to GraphViz Instructions for Email graph server. Overview paper on GraphViz

Problem Links ● ●

Drawing Graphs Nicely (7) Drawing Trees (7)

About the Book Send us Mail Go to Main Page This page last modified on Feb 10, 1997.

file:///E|/WEBSITE/IMPLEMEN/NORTH/IMPLEMEN.HTM [19/1/2003 1:40:45]

Mike Trick's Graph Coloring Resources

Mike Trick's Graph Coloring Resources Michael Trick's WWW page on graph coloring, http://mat.gsia.cmu.edu/COLOR/color.html, provides a nice overview of applications of graph coloring, an annotated bibliography, and a collection of over 70 graph coloring instances arising in applications such as register allocation and printed circuit board testing. Finally, it contains a C langauge implementation of an exact coloring algorithm, DSATUR. ● ● ●

Link to Trick's Graph Coloring WWW page Link to Trick's Operations Research WWW page Download Files (local site)

Problem Links ● ●

Vertex Coloring (7) Edge Coloring (4)

About the Book Send us Mail Go to Main Page This page last modified on Feb 23, 1996.

file:///E|/WEBSITE/IMPLEMEN/TRICK/IMPLEMEN.HTM [19/1/2003 1:40:46]

Joe Culberson's Graph Coloring Resources

Joe Culberson's Graph Coloring Resources Joseph C. Culberson's WWW page on graph coloring, http://web.cs.ualberta.ca/~joe/Coloring/, provides an extensive bibliography and a collection of programs to generate hard graph coloring instances. ● ● ●

Link to Culberson's Graph Coloring WWW page Link to Culberson's home page Download Files (local site)

Problem Links ● ●

Vertex Coloring (6) Edge Coloring (4)

About the Book Send us Mail Go to Main Page This page last modified on Feb 23, 1996.

file:///E|/WEBSITE/IMPLEMEN/CULBERSO/IMPLEMEN.HTM [19/1/2003 1:40:46]

Frank Ruskey's Combinatorial Generation Resources

Frank Ruskey's Combinatorial Generation Resources An exciting WWW site developed by Frank Ruskey of the University of Victoria contains a wealth of material on generating combinatorial objects of different types, including permutations, subsets, partitions, and certain graphs. Specifically, there is an interactive interface which lets you specify which type of objects you would like to construct, and quickly returns the objects to you. For certain types of objects, programs which generate them are available by ftp. It is well worth checking this out at http://www-csc.uvic.ca/home/fruskey/cgi-bin/html/main.html His home page contains pointers to other useful material on generating combinatorial objects, along with interesting graphics.

● ●

Link to Ruskey's Combinatorial Generation WWW page Link to Ruskey's home page -- interesting!

Problem Links ● ● ● ●

Generating Permutations (8) Generating Graphs (7) Generating Partitions (7) Generating Subsets (7)

About the Book Send us Mail Go to Main Page This page last modified on Feb 10, 1997.

file:///E|/WEBSITE/IMPLEMEN/RUSKEY/IMPLEMEN.HTM [19/1/2003 1:40:47]

Triangle: A Two-Dimensional Quality Mesh Generator

Triangle: A Two-Dimensional Quality Mesh Generator Triangle, by Jonathan Shewchuk of Carnegie-Mellon University, is a C language code which generates Delaunay triangulations, constrained Delaunay triangulations (forced to have certain edges), and quality conforming Delaunay triangulations (which avoid small angles by inserting extra points). It has been widely used for finite element analysis and other applications, and is fast and robust. Triangle would be the first thing I would try if I was looking for two-dimensional triangulation code. Although Triangle is available at http://www.cs.cmu.edu/~quake/triangle.html, it is copyrighted by the author and may not be sold or included in commercial products without a license. Also included is a package by Shewchuk for robust computation of geometric predicates using exact arithmetic. ● ● ● ●

Triangle home page Netlib source directory Robust Geometric Primatives Download Files (local site)

Problem Links ●

Triangulation (9)

About the Book Send us Mail Go to Main Page This page last modified on Sep 26, 1996.

file:///E|/WEBSITE/IMPLEMEN/TRIANGLE/IMPLEMEN.HTM [19/1/2003 1:40:48]

Arrange - maintainance of arrangements with point location

Arrange - maintainance of arrangements with point location Arrange is a package written in C by Michael Goldwasser of Stanford University for maintaining arrangements of polygons in either the plane or on the sphere. Polygons may be degenerate, and hence represent arrangements of lines. A randomized incremental construction algorithm is used, and efficient point location on the arrangement supported. Polygons may be inserted but not deleted from the arrangement, and arrangements of several thousand vertices and edges can be constructed in a few seconds. Arrange is available from ftp://flamingo.stanford.edu/pub/wass/arrangement. ● ● ●

Arrange ftp source directory Michael Goldwasser's home page Download Files (local site)

Problem Links ● ●

Maintaining Line Arrangements (9) Point Location (6)

About the Book Send us Mail Go to Main Page This page last modified on Apr 24, 1996.

file:///E|/WEBSITE/IMPLEMEN/ARRANGE/IMPLEMEN.HTM [19/1/2003 1:40:48]

Linprog -- low dimensional linear programming

Linprog -- low dimensional linear programming For low-dimensional linear programming problems, computational geometry flavored algorithms can outperform more general LP codes. An implementation of Seidel's randomized incremental LP algorithm in C by Mike Hohmeyer at ftp://icemcfd.com/pub/linprog.a. This program also handles rational objective functions, so with some cleverness you can get polytope separation distance, linear programming on a sphere, etc. ● ● ●

Original FTP site Linear programming FAQ . Download Files (local site)

Problem Links ●

Linear Programming (4)

About the Book Send us Mail Go to Main Page This page last modified on Apr 29, 1996.

file:///E|/WEBSITE/IMPLEMEN/LINPROG/IMPLEMEN.HTM [19/1/2003 1:40:49]

LP_SOLVE: Linear Programming Code

LP_SOLVE: Linear Programming Code The non-commercial linear programming code of choice appears to be lp_solve , written in ANSI C by Michel Berkelaar, who claims to have solved problems as large as 30,000 variables and 50,000 constraints. Lp_solve can also handle (smaller) integer and mixed-integer problems. It is available by anonymous ftp from ftp://ftp.es.ele.tue.nl/pub/lp_solve, but is not in the public domain. A user community for lp_solve exists, which has ported it to a variety of different platforms. ● ● ● ●

FTP site for lp_solve Mail to author Michel Berkelaar Linear Programming FAQ Download Files (local site)

Problem Links ●

Linear Programming (9)

About the Book Send us Mail Go to Main Page This page last modified on Apr 29, 1996.

file:///E|/WEBSITE/IMPLEMEN/LPSOLVE/IMPLEMEN.HTM [19/1/2003 1:40:50]

PARI - Package for Number Theory

PARI - Package for Number Theory PARI is a system capable of handling complex number-theoretic and algebraic problems. The predefined types are integers (up to 300000 decimal digits), reals (same for the relative precision), elements of Z/nZ, rational numbers, complex numbers, p-adic numbers, quadratic numbers, polynomials, power series, algebraic extensions, rational functions, binary quadratic forms, vectors, matrices. More than 200 special predefined functions, arithmetic or transcendental, are implemented, in addition to the usual arithmetic operations, which can be used without paying attention to the type of the object. The source uses more than 32000 lines of code, mainly in C. PARI can be used as a library, but possesses also a powerful calculator mode which gives instant access to all the types and functions. In this mode, one can write programs with a simple syntax a little similar to C, but taking also LISP-like constructs. The main advantage of PARI is its speed. On a Unix platform, it is between 5 to 100 times faster than Maple or Mathematica, depending on the applications. Also it is specially tailored for use by number theorists, hence contains a large number of predefined number-theoretical functions not found in other systems. It can of course profitably be used by other people as well. Finally, it is free. Currently, PARI is available for PC, Amiga, Macintosh, and most Unix platforms by anonymous ftp at megrez.ceremab.u-bordeaux.fr in the directory pub/pari . The current version of PARI is 1.39 (Jan. 1995). The email address of the PARI group is [email protected] .

● ● ●

PARI description Original FTP site in Bordeaux Download Files (local site)

Problem Links ● ●

Factoring and Primality Testing (9) Arbitrary Precision Arithmetic (9)

About the Book

file:///E|/WEBSITE/IMPLEMEN/PARI/IMPLEMEN.HTM (1 of 2) [19/1/2003 1:40:50]

PARI - Package for Number Theory

Send us Mail Go to Main Page This page last modified on May 7, 1996.

file:///E|/WEBSITE/IMPLEMEN/PARI/IMPLEMEN.HTM (2 of 2) [19/1/2003 1:40:50]

file:///E|/WEBSITE/IMPLEMEN/GRAPHED/IMPLEMEN.HTM

GraphEd -- Graph Editor and Layout Program GraphEd is a graph editor by Michael Himsolt, which provides graph layout algorithms and an interface for application modules. It provides a wide variety of graph drawing algorithms -- including spring embeddings and special algorithms for drawing trees, DAGs, and planar graphs. GraphEd is probably the best place to start if you are looking for graph drawing software or algorithms. GraphEd is written in ANSI C, for UNIX systems running X-windows. GraphEd is available for non-commercial use via anonymous ftp from `ftp.uni-passau.de' (132.231.1.10) in `/pub/local/graphed' or from our local site. In our experience, it does crash from time to time, but it is a large and sophisticated piece of software which has been fairly widely and successfully used.

● ● ● ●

Original FTP site Graphlet WWW page (successor system to GraphEd) Graph Drawing Tools Page Download Files (local site)

Problem Links ● ● ● ● ● ● ●

Drawing Graphs Nicely (9) Drawing Trees (9) Graph Data Structures (8) Planarity Detection and Embedding (8) Connected Components (4) Triangulation (4) Network Flow (3)

About the Book Send us Mail Go to Main Page This page last modified on Apr 18, 1997.

file:///E|/WEBSITE/IMPLEMEN/GRAPHED/IMPLEMEN.HTM [19/1/2003 1:40:52]

TSP solvers

TSP solvers tsp-solve is a C++ code by Chad Hurwitz and Robert Craig with and provides both heuristic and optimal solutions. Geometric problems of size up to 100 points are managable. It is available by emailing Chad Hurwitz at [email protected], and below on the algorithm repository. A heuristic Euclidean TSP solver in C due to Lionnel Maugis is available from http://www.cenaath.cena.dgac.fr/~maugis/tsp.shar . ● ●

Lionnel Maugis's TSP_heu Download Files (local site)

Problem Links ●

Traveling Salesman Problem (8)

About the Book Send us Mail Go to Main Page This page last modified on Aug 9, 1996.

file:///E|/WEBSITE/IMPLEMEN/TSP/IMPLEMEN.HTM [19/1/2003 1:40:52]

FFTPACK -- Fourier Transform Library

FFTPACK -- Fourier Transform Library

FFTPACK is a package of Fortran subprograms for the fast Fourier transform of periodic and other symmetric sequences, written by P. Swarztrauber. It includes complex, real, sine, cosine, and quarterwave transforms. A C language translation of the main routines is also provided.

● ●

Original distribution from Netlib Download Files (local site)

Problem Links ●

Discrete Fourier Transform (10)

About the Book Send us Mail Go to Main Page This page last modified on Aug 13, 1996.

file:///E|/WEBSITE/IMPLEMEN/FFTPACK/IMPLEMEN.HTM [19/1/2003 1:40:53]

PHYLIP -- inferring phylogenic trees

PHYLIP -- inferring phylogenic trees PHYLIP is an extensive and widely-used package of programs for inferring phylogenies. It contains over thirty different algorithms for constructing phylogenic trees. Although many of them are designed to work with molecular sequence data, several general methods accept arbitrary distance matrices as input. With versions written in C and Pascal, and binaries for Windows, DOS, Macintosh, and PowerMac, it is available on the WWW from http://evolution.genetics.washington.edu/phylip.html. ●

PHYLIP distribution site

Problem Links ●

Steiner Tree (7)

About the Book Send us Mail Go to Main Page This page last modified on May 2, 1997.

file:///E|/WEBSITE/IMPLEMEN/PHYLIP/IMPLEMEN.HTM [19/1/2003 1:40:53]

Salowe's Rectilinear Steiner trees

Salowe's Rectilinear Steiner trees Salowe and Warme developed a program for computing exact rectilinear Steiner minimal trees, which should be capable of handling up to 30 points routinely. It is available by anonymous ftp from ftp.cs.virginia.edu in pub/french/salowe/newsteiner.tar.Z. The program is described in the paper "ThirtyFive-Point Rectilinear Steiner Minimal Trees in a Day", in Networks: An International Journal , volume 25, 1995. Also available is another Steiner tree program from Virgina, by Gabriel Robins and Tong-Tong Zhang. This one produces a good heuristic solution.

● ● ●

Original FTP site Send Jeff Salowe mail Download Files (local site)

Problem Links ●

Steiner Tree (8)

About the Book Send us Mail Go to Main Page This page last modified on Sep 19, 1996.

file:///E|/WEBSITE/IMPLEMEN/SALOWE/IMPLEMEN.HTM [19/1/2003 1:40:54]

Skeletonization Software (2-D)

Skeletonization Software (2-D) MAT is a medial-axis transform code designed for 2-D skeletonization of binary images, written by Robert Ogniewicz and available from http://hrl.harvard.edu/people/postdocs/rlo/rlo.dir/rlo-soft.html. MAT accepts a variety of different input formats, including polygonal representations. This seem to be a solidly built program, and should be your first stop on seeking a routine for thinning. ● ● ●

Skeletonation WWW page Send mail to R. Ogniewicz Download Files (local site)

Problem Links ● ●

Medial-Axis Transformation (9) Simplifying Polygons (5)

About the Book Send us Mail Go to Main Page This page last modified on Oct 4, 1996.

file:///E|/WEBSITE/IMPLEMEN/SKELETON/IMPLEMEN.HTM [19/1/2003 1:40:55]

SNNS - Stuttgart Neural Network Simulator

SNNS - Stuttgart Neural Network Simulator The "Stuttgart Neural Network Simulator" from the University of Stuttgart, Germany supports many types of networks and training algorithms, as well as sophisticated graphical visualization tools under X11. It has been ported to many flavors of UNIX. It is available for ftp from ftp.informatik.unistuttgart.de [129.69.211.2] in directory /pub/SNNS as SNNSv4.1.tar.gz (1.4 MB, Source code) and SNNSv4.1.Manual.ps.gz (1 MB, Documentation). It may be best to first have a look at the file SNNSv4.1.Readme. More information can be found in the WWW under http://www.informatik.unistuttgart.de/ipvr/bv/projekte/snns/snns.html ● ● ●

Original distribution site Neural Networks FAQ, with pointers to other software Download Files (local site)

Problem Links ●

Shape Similarity (7)

About the Book Send us Mail Go to Main Page This page last modified on Oct 9, 1996.

file:///E|/WEBSITE/IMPLEMEN/SNNS/IMPLEMEN.HTM [19/1/2003 1:40:55]

agrep - Approximate General Regular Expression Pattern Matcher

agrep - Approximate General Regular Expression Pattern Matcher The best available tools for approximate pattern matching are glimpse and agrep, developed by Manber and Wu at the University of Arizona, and available from http://glimpse.cs.arizona.edu/. Glimpse is a tool for building and using an index for searching through file systems, while agrep (approximate general regular expression pattern matcher) is a tool supporting text search with spelling errors. Both programs are widely used and respected. ● ●

Official Glimpse and agrep distribution Download Files (local site)

Problem Links ●

Approximate String Matching (10)

About the Book Send us Mail Go to Main Page This page last modified on Oct 13, 1996.

file:///E|/WEBSITE/IMPLEMEN/AGREP/IMPLEMEN.HTM [19/1/2003 1:40:56]

HT/DIG -- image compression codes

HT/DIG -- image compression codes An alternate WWW text search engine is {\em ht://Dig} from Andrew Scherpbier, which also contains implementations of Soundex and Metaphone. It is available from http://htdig.sdsu.edu/, and released under the GNU general public licence. ● ●

Source ht://dig distribution Download Files (local site)

Problem Links ●

Approximate String Matching (7)

About the Book Send us Mail Go to Main Page This page last modified on Oct 13, 1996.

file:///E|/WEBSITE/IMPLEMEN/HTDIG/IMPLEMEN.HTM [19/1/2003 1:40:57]

CAP -- Contig Assembly Program

CAP -- Contig Assembly Program CAP (Contig Assembly Program) by Xiaoqiu Huang is a C language program supporting DNA shotgun sequencing, by finding the shortest common superstring of a set of fragments. As to performance, CAP took 4 hours to assemble 1015 fragments of a total of 252,000 characters on a Sun SPARCstation SLC. Certain parameters will need to be tweaked to make it accommodate non-DNA data. It is available by anonymous ftp from cs.mtu.edu in the pub/huang directory. MAP (Multiple Alignment Program) also by Xiaoqiu Huang is a C language program computes a multiple global alignment of sequences using an iterative pairwise method. It is also available by anonymous ftp from cs.mtu.edu in the pub/huang directory. ● ●

Download Files (local site) Author's homepage with original sources

Problem Links ● ●

Longest Common Substring (8) Shortest Common Superstring (8)

About the Book Send us Mail Go to Main Page This page last modified on Oct 16, 1996.

file:///E|/WEBSITE/IMPLEMEN/CAP/IMPLEMEN.HTM [19/1/2003 1:40:57]

Shape similarity testing via turning functions

Shape similarity testing via turning functions This is an implementation in C by Eugene K. Ressler of the turning function metric for comparing polygonal shapes developed by Arkin, Chew, Huttenlocher, Kedem, and Mitchell. It expands a little on the cited reference to achieve O(n) space and O(mn log n) run time. This source may be freely distributed and used for non-commercial purposes, so long as this comment is attached to any code copied or derived from it. ●

Download Files (local site)

Problem Links ●

Shape Similarity (6)

About the Book Send us Mail Go to Main Page This page last modified on Oct 10, 1996.

file:///E|/WEBSITE/IMPLEMEN/TURN/IMPLEMEN.HTM [19/1/2003 1:40:58]

NAUTY -- Graph Isomorphism

NAUTY -- Graph Isomorphism The world's fastest isomorphism testing program is Nauty, by Brendan D. McKay. Nauty (No AUTomorphisms, Yes?) is a set of very efficient C language procedures for determining the automorphism group of a vertex-colored graph. It provides this information in the form of a set of generators, the size of group, and the orbits of the group. Nauty is also able to produce a canonicallylabeled isomorph of the graph, to assist in isomorphism testing. It was the basis of the first program to generate all the 11-vertex graphs without isomorphs, and can test most graphs of less than 100 vertices in well under a second. Nauty has been successfully ported to a variety of operating systems and C compilers. It may be obtained from http://cs.anu.edu.au/people/bdm/. It is free for educational and research applications, but for commercial use contact the author at [email protected]. ● ●

Download Files (local site)

Problem Links ●

Graph Isomorphism (10)

About the Book Send us Mail Go to Main Page This page last modified on Apr 23, 1997.

file:///E|/WEBSITE/IMPLEMEN/NAUTY/IMPLEMEN.HTM [19/1/2003 1:40:59]

POSIT - Propositional Satisfiability Testbed

POSIT - Propositional Satisfiability Testbed The propositional satisfiability tester POSIT, by Jon W. Freeman, is based on a highly optimized version of the Davis-Putnum procedure. It is available by anonymous ftp from ftp.cis.upenn.edu in /pub/freeman/posit-1.0.tar.Z. ● ●

POSIT (primary FTP site) Download Files (local site)

Problem Links ●

Satisfiability (8)

About the Book Send us Mail Go to Main Page This page last modified on Apr 23, 1997.

file:///E|/WEBSITE/IMPLEMEN/POSIT/IMPLEMEN.HTM [19/1/2003 1:40:59]

BIPM -- Bipartite Matching Codes

BIPM -- Bipartite Matching Codes This directory contains several programs that implement algorithms for bipartite matching in unweighted graphs. Except where noted, all programs were written by Joao Carlos Setubal ([email protected]), who also wrote this documentation. You can use and modify the codes freely, but please acknowledge authorship. If a bibliographic reference is needed, use the following: J. C. Setubal, Sequential and Parallel Experimental Results with Bipartite Matching Algorithms. Technical Report IC-96-09, Institute of Computing, State University of Campinas (Brazil), 1996. A postscript file containing this tech report is in directory techreport, compressed with gzip. It will be easier to understand what each code does if you read the tech report first.

● ●

Original distribution Download Files (local site)

Problem Links ●

Matching (8)

About the Book Send us Mail Go to Main Page This page last modified on Apr 23, 1997.

file:///E|/WEBSITE/IMPLEMEN/BIPM/IMPLEMEN.HTM [19/1/2003 1:41:00]

GEOMPACK - triangulation and convex decomposition codes

GEOMPACK - triangulation and convex decomposition codes GEOMPACK is a suite of Fortran 77 codes by Barry Joe of the University of Alberta, for 2- and 3dimensional triangulation and convex decomposition problems. In particular, it does both Delaunay triangulation and convex decompositions of polygonal and polyhedral regions, as well as arbitrary dimensional Delaunay triangulations. These codes are well respected, and worth checking out. They can be obtained from ftp://ftp.cs.ualberta.ca/pub/geompack. Papers describing the algorithms used for each program are available at the ftp site. The GEOMPACK software package is copyrighted and made available here for noncommercial use only. If you want to include any GEOMPACK routines in a commercial software product, please contact Barry Joe about obtaining a license.

● ● ●

FTP site for GEOMPACK Barry Joe's home page Download Files (local site)

Problem Links ● ●

Polygon Partitioning (8) Triangulation (8)

About the Book Send us Mail Go to Main Page This page last modified on Apr 24, 1996.

file:///E|/WEBSITE/IMPLEMEN/GEOMPACK/IMPLEMEN.HTM [19/1/2003 1:41:01]

LAPACK and LINPACK -- Linear Algebra PACKages

LAPACK and LINPACK -- Linear Algebra PACKages LAPACK provides routines for solving systems of simultaneous linear equations, least-squares solutions of linear systems of equations, eigenvalue problems, and singular value problems. The associated matrix factorizations (LU, Cholesky, QR, SVD, Schur, generalized Schur) are also provided, as are related computations such as reordering of the Schur factorizations and estimating condition numbers. Dense and banded matrices are handled, but not general sparse matrices. In all areas, similar functionality is provided for real and complex matrices, in both single and double precision. LINPACK, also provided in the distrib directory, is an older linear algebra package, of which LAPACK is its successor. We have included LINPACK because it more obviously solves determinants than LAPACK.

● ● ●

Source LAPACK distribution Source LINPACK distribution Download Files (local site)

Problem Links ●

Solving Linear Equations (10)

About the Book Send us Mail Go to Main Page This page last modified on Aug 13, 1996.

file:///E|/WEBSITE/IMPLEMEN/LAPACK/IMPLEMEN.HTM [19/1/2003 1:41:01]

Mathematica -- Assorted Routines

Mathematica -- Assorted Routines These files contain implementations of interesting algorithms in the Mathematica programming language. A large collection of Mathematica programs is MathSource, maintained by Wolfram Research at http://www.wri.com/mathsource/. Many of the files below are from MathSource, while others were obtained directly from the authors. ●

Strassen's Matrix Multiplication Algorithm -- An implementation offered ``without comment or promise of efficiency'', by Stan Wagon of Macalester College, [email protected] . in distrib/strassen.m

Combinatorica, a library of 230 algorithms for combinatorics and graph theory written in Mathematica, has a separate entry in this algorithm repository.

● ●

Download Files (local site) MathSource at Wolfram Research

Problem Links ●

Matrix Multiplication (6)

About the Book Send us Mail Go to Main Page This page last modified on Aug 10, 1996.

file:///E|/WEBSITE/IMPLEMEN/MATH/IMPLEMEN.HTM [19/1/2003 1:41:02]

User Comments

User Comments Archive ● ●

([email protected]):Welcome to the User Comments Archive ([email protected]):Test of user comments

About the Book Send us Mail Go to Main Page

file:///E|/WEBSITE/FEEDBACK/COMMENTS/COMMENTS.HTM [19/1/2003 1:41:03]

The Algorithm Design Manual

The Algorithm Design Manual ● ● ● ●

Overview Contents About the Author To Order

Author Steven S. Skiena Department of Computer Science State University of New York at Stony Brook

Overview http://www.telospub.com/catalog/COMPUTERSCI/Algorithm.html (1 of 5) [19/1/2003 2:00:54]

The Algorithm Design Manual

Most professional programmers are not well equipped to tackle algorithm design problems. "The Algorithm Design Manual" by Steve Skiena of SUNY Stony Brook, is uniquely designed to provide access to combinatorial algorithms technology for computer professionals and students. Skiena has taken some of the "mystery" out of finding the right algorithm for the job, by drawing heavily on the author's own real-world experiences. Thus the work takes on a very practical character, as reflected in its title: "Manual". The Book is divided into two parts, the first being a general guide to techniques for the design and analysis of computer algorithms. The second is a reference section, comprising the catalog of algorithmic resources, implementations, and an extensive bibliography. The primary audience for this book/CD-ROM combination is the working professional who uses algorithms on a regular or occasional basis and has need for a handy reference. A major feature of this book is the inclusion of a complete "catalog" of important algorithmic problems. By browsing this catalog, readers can quickly identify what the problem they have encountered is called, what is known about it, and how they should proceed if they need to solve it. Nothing like this catalog exists in the computing literature for general computer algorithms. This work can also readily be used as a textbook or course supplement in standard courses on algorithm design. Pedagogic features include pen-and paper exercises, "team projects", independent student projects, "take home" lessons (goals) at the beginning of chapters. Other teaching and learning aids reside on the accompanying CD-ROM. The multiplatiform CD-ROM contains a full hypertext version of the book, with a comprehensive online index, and all of the code/algorithms residing on the author's web site at Stony Brook in the "Algorithm Repository" there: www.cs.sunysb.edu/~algorith/. URLs for all cited implementations mirroring the Stony Brook web site and algorthim repository are included. Also included on the CD-ROM are 10 hours of audio lectures presented by the author, and series of slides which instructors can use to help teach their courses. There is additional useful information and updates which are available through accessing the author's web site.

Table of Contents Part I Techniques Chapter 1 Introduction to Algorithms 1.1 Correctness and Efficiency 1.2 Expressing Algorithms 1.3 Keeping Score http://www.telospub.com/catalog/COMPUTERSCI/Algorithm.html (2 of 5) [19/1/2003 2:00:54]

The Algorithm Design Manual

1.4 The Big Oh Notation 1.5 Growth Rates 1.6 Logarithms 1.7 Modeling the Problem 1.8 About the War Stories 1.9 War Story: Psychic Modeling 1.10 Exercises Chapter 2 Data Structures and Sorting 2.1 Fundamental Data Types 2.2 Specialized Data Structures 2.3 Sorting 2.4 Applications of Sorting 2.5 Approaches to Sorting 2.6 War Story: Stripping Triangulations 2.7 War Story: Mystery of the Pyramids 2.8 War Story: String em Up 2.9 Exercises Chapter 3 Breaking Problems Down 3.1 Dynamic Programming 3.2 Limitations of Dynamic Programming 3.3 War Story: Evolution of the Lobster 3.4 War Story: What s Past is Prolog 3.5 War Story: Text Compression for Bar Codes 3.6 Divide and Conquer 3.7 Exercises Chapter 4 Graph Algorithms 4.1 The Friendship Graph 4.2 Data Structures for Graphs 4.3 War Story: Getting the Graph 4.4 Traversing a Graph 4.5 Applications of Graph Traversal 4.6 Modeling Graph Problems 4.7 Minimum Spanning Trees 4.8 Shortest Paths 4.9 War Story: Nothing but Nets 4.10 War Story: Dialing for Documents 4.11 Exercises Chapter 5 Combinatorial Search and Heuristic Methods 5.1 Backtracking 5.2 Search Pruning 5.3 Bandwidth Minimization 5.4War Story: Covering Chessboards 5.5 Heuristic Methods http://www.telospub.com/catalog/COMPUTERSCI/Algorithm.html (3 of 5) [19/1/2003 2:00:54]

The Algorithm Design Manual

5.6 War Story: Annealing Arrays 5.7 Parallel Algorithms 5.8 War Story: Going Nowhere Fast 5.9 Exercises Chapter 6 Intractable Problems and Approximations 6.1 Problems and Reductions 6.2 Simple Reductions 6.3 Satisfiability 6.4 Difficult Reductions 6.5 Other NP-Complete Problems 6.6 The Art of Proving Hardness 6.7 War Story: Hard Against the Clock 6.8 Approximation Algorithms 6.9 Exercises Chapter 7 How To Design Algorithms Part II Resources Chapter 8 A Catalog of Algorithmic Problems 8.1 Data Structures 8.2 Numerical Problems 8.3 Combinatorial Problems 8.4 Graph Problems - polynomial-time problems 8.5 Graph Problems - hard problems 8.6 Computational Geometry 8.7 Set and String Problems Chapter 9 Algorithmic Resources 9.1 Software Systems 9.2 Data Sources 9.3 Textbooks 9.4 On-Line Resources 9.5 Professional Consulting Services 1997/496 pages/Hardcover ISBN 0-387-94860-0 Back to top

About the Author http://www.telospub.com/catalog/COMPUTERSCI/Algorithm.html (4 of 5) [19/1/2003 2:00:54]

The Algorithm Design Manual

Steve Skiena is currently an Associate Professor of Computer Science at the State University of New York, Stony Brook where he has been since 1988. Back to top

To Order ❍

Order a Book Online ❍

❍

Detailed Ordering Information To order by regular mail, send orders to: Springer-Verlag New York, Inc. PO Box 2485 Secaucus, NJ 07096-2485

❍

To order by phone, call: 1-800-777-4643

❍

To order by fax: 201-348-4505

Back to top

Back to TELOS Catalog

Back to TELOS Home Page

http://www.telospub.com/catalog/COMPUTERSCI/Algorithm.html (5 of 5) [19/1/2003 2:00:54]